Infrared vision sensing detection method and device for rocking arc bottoming weld penetration

Abstract

The invention discloses an infrared vision sensing detection method and an infrared vision sensing detection device for rocking arc bottoming weld penetration. The infrared vision sensing detection method comprises the following steps: intercepting left and right image small windows by using double windows, or intercepting a molten pool tail image small window by using a single window; taking edges between the highest point and the maximum curvature point in the left and right images as left and right feature edges, or searching first extreme points of longitudinal coordinates of the edges from left to right and from right to left in a part above half of the height of the edge image at the tail of the molten pool; searching an edge curvature maximum point from the left extreme point to the left and from the right extreme point to the right, and intercepting an edge between the extreme point and the corresponding edge curvature maximum point as a left part and a right part of a feature edge; fitting the feature edge by adopting a quadratic curve in the global image coordinate system to obtain a curvilinear equation and a quadratic term coefficient thereof, and taking the quadratic term coefficient as a feature parameter detection value; the fusion penetration state information is obtained by detecting the characteristics of the tail of the rocking arc backing weld molten pool, and high fusion penetration sensing detection precision is achieved.

Description

Technical Field

[0001] The invention relates to the technical field of welding, and in particular to an infrared visual sensing method and device for real-time detection of the penetration state of a shaking arc backing weld. Background Art

[0002] In the manufacturing of large-scale equipment such as ships, pipelines, and pressure vessels, the application of automated welding technology is becoming more and more extensive, and the demand for welding of medium and thick plates is also increasing. For some medium and thick plate structural parts that are not easy to flip or cannot be flipped or cannot be welded on the reverse side, it is often necessary to use a single-sided welding and double-sided forming backing welding process technology to carry out the first layer welding of the groove (commonly known as backing welding). For the backing welding of structural parts, the penetration quality of the backing weld is an important indicator for evaluating the quality of the backing welding process. During manual arc welding, the welding operator observes the changes in the front morphology of the molten pool with the naked eye to judge the penetration state of the weld, and adjusts the welding process parameters in real time to avoid defects such as welding penetration and incomplete welding. Therefore, the front appearance morphological characteristics of the backing weld pool can reflect the penetration state of the backing weld. In the multi-layer and multi-pass welding process of medium and thick plates, the detection and control of the penetration state (especially when backing welding) are the key points and technical difficulties in realizing the automation and intelligent welding of medium and thick plates. Among them, the first technical difficulty that needs to be solved is how to identify the penetration state of the weld by detecting the front morphological characteristics of the molten pool.

[0003] The Chinese patent application number 202011556910.X, entitled "A method for melt penetration identification based on bidirectional melt pool geometry and texture feature fusion", discloses a method of using a bidirectional synchronous visual sensing system to obtain clear and synchronous melt pool images from the upper rear and upper side of the melt pool, and to establish an RBF melt penetration identification model by extracting the geometric features of the upper rear melt pool image and the texture feature information of the upper side melt pool image to predict the melt penetration state. The disadvantages of this method are: (1) two cameras are required to synchronously capture the melt pool image, and the system is complex and costly; (2) arc light, smoke and other interference issues are not considered during image processing, and the anti-interference ability is poor. In addition, the global image processing time is relatively long, which increases the subsequent melt penetration control speed requirements. The document entitled "VISION-BASED TIG OVERHEAD WELDING ROOT PENALTY STATE RECOGNITION METHOD FOR TIG OVERHEAD WELDING OF THICK PLATE GROOVE" with Chinese application number 202410095608.0 discloses a method for synchronously collecting double-sided images of the molten pool from the front and back sides of the molten pool, extracting the size features of the back molten pool, and establishing a penetration recognition model by deep learning the front molten pool features under different penetration states to obtain penetration state information. The disadvantages of this method are: (1) Two industrial cameras are required, and the molten pool images need to be collected from the front and back sides of the molten pool at the same time, which significantly increases the complexity of the system and is not suitable for situations where the back side image of the molten pool cannot be collected; (2) It is necessary to collect welding images under various penetration states in advance, with a large number of samples, and the global image is processed, which increases the amount of data and time for model training. Summary of the invention

[0004] In view of the shortcomings of the prior art such as complex system of the base weld penetration identification device, low efficiency of the penetration detection method and poor anti-interference ability, the present invention proposes an infrared visual sensing detection device with simple system structure, good real-time performance, strong environmental adaptability and anti-interference ability, which is suitable for the penetration of the shaking arc base weld. The base weld penetration status information is obtained by detecting the tail characteristics of the shaking arc base weld molten pool.

[0005] In order to achieve the above-mentioned purpose of the invention, the technical solution adopted by the infrared visual sensing detection method of the swing arc bottom weld penetration proposed by the present invention comprises the following steps:

[0006] ① Move the infrared camera and the shaking arc welding torch to the front of the groove of the workpiece to be welded and collect the welding area image, and send the welding area image to the computer processing module;

[0007] ② The computer processing module processes the welding area image to obtain the maximum circumscribed rectangle of the molten pool contour, extracts the centroid coordinate value, height value and width value of the maximum circumscribed rectangle, and uses a double-window interception method to intercept the left and right image windows at the tail of the molten pool, or uses a single-window interception method to intercept the image window at the tail of the molten pool;

[0008] ③ Processing the small window images of the left and right sides of the tail of the molten pool or the small window images intercepted by the small window images of the tail of the molten pool to obtain the left and right edge images of the tail of the molten pool or the edge images of the tail of the molten pool;

[0009] For the left and right edge images corresponding to the left and right image small windows of the tail of the molten pool, respectively extract the highest points of the left and right edges and the maximum points of the edge curvature correspondingly searched to the left and right from the highest points, and intercept the edges between the highest points and the maximum curvature points in the corresponding left and right edge images as the left and right feature edges;

[0010] For the molten pool tail edge image corresponding to the molten pool tail image small window, search for the first extreme point of the ordinate of the molten pool tail edge from the left side of the image to the right and from the right side of the image to the left in the portion above the half of the height of the edge image, and search for the maximum point of edge curvature from the extreme point searched to the right to the left and from the extreme point searched to the left to the right in the molten pool tail edge image, and intercept the edge between the extreme point searched to the right and the corresponding maximum point of edge curvature searched to the left, and the extreme point searched to the left and the corresponding maximum point of edge curvature searched to the right as the left and right parts of the characteristic edge of the molten pool tail, which are collectively used as the characteristic edge of the molten pool tail;

[0011] ④ Simultaneously map the left and right feature edges to the global image coordinate system xOy, or map the feature edge of the tail of the molten pool to the global image coordinate system xOy, and use the quadratic curve fitting method to perform quadratic curve fitting calculation on the left and right feature edges or the feature edge of the tail of the molten pool, obtain the fitting edge of the tail of the molten pool, the curve equation of the corresponding fitting edge in the global image coordinate system xOy and the quadratic term coefficient in the quadratic curve equation, and use the quadratic term coefficient as the current molten pool tail shape feature parameter detection value a i ;

[0012] ⑤ By detecting the characteristic parameter value a of the tail shape of the current molten pool i Filter to obtain the sampling value a of the shape characteristic parameter of the current molten pool tail si , determine the sampling value a of the current molten pool tail shape characteristic parameter si , if a si ≤a min The current weld is in the weld-through state. min <a si <a max The current weld is in the penetration state. If a si ≥a max The current weld is in an incomplete penetration state, a min is the criterion for the penetration of the swing arc backing weld, a max It is the criterion for judging whether the shaking arc base weld is not fully penetrated.

[0013] Further, in step ②, the dual-window interception method is: first extract the coordinate value of the end point of the welding wire from the inside of the maximum circumscribed rectangle, and then adjust the coordinate values ​​of the positioning points of the left and right image windows of the tail of the molten pool according to the changes in the coordinate value, height and width of the center of gravity of the maximum circumscribed rectangle, and adjust the height and width of the left and right image windows of the tail of the molten pool according to the changes in the coordinate value, height and width of the center of gravity of the maximum circumscribed rectangle and the coordinate value of the end point of the welding wire; the single-window interception method is: adjust the coordinate value of the positioning point of the small window of the image of the tail of the molten pool according to the changes in the coordinate value, height and width of the center of gravity of the maximum circumscribed rectangle, and adjust the height and width of the small window for intercepting the image of the tail of the molten pool according to the changes in the height and width of the maximum circumscribed rectangle.

[0014] Furthermore, the method for extracting the coordinates of the end point of the welding wire is: on the global binary image of the molten pool, by searching for a pixel point with a grayscale value of 0 and a minimum vertical coordinate value in an 80pixels×80pixels area centered on the centroid of the largest circumscribed rectangle, and using the coordinates of the pixel point as the coordinates of the end point of the welding wire.

[0015] Furthermore, the method for determining the small image windows on the left and right sides of the tail of the molten pool is: determine the coordinate values ​​of the positioning points of the small image windows on the left and right sides of the tail of the molten pool according to the centroid coordinate value, height value and width value of the maximum circumscribed rectangle, and make the vertical coordinate values ​​of the positioning points of the small image windows on the left and right sides of the tail of the molten pool equal.

[0016] Furthermore, when the horizontal coordinate value of the end point of the welding wire is less than or equal to the horizontal coordinate value of the center of gravity of the maximum circumscribed rectangle, the width of the small image window on the left side of the tail of the molten pool is w i / 2-|x Pi -x Ei |、Height value is h i / 2+|y Pi -y Ei |, the width of the small window of the image intercepted on the right side of the molten pool tail is w i / 2+|x Pi -x Ei |、Height value is h i / 2-|y Pi -y Ei |; When the horizontal coordinate value of the end point of the welding wire is greater than the horizontal coordinate value of the center of gravity of the maximum circumscribed rectangle, the width of the small window for capturing the image on the left side of the tail of the molten pool is w i / 2+|x Pi -x Ei |、Height value is h i / 2-|y Pi -y Ei |, the width of the small window of the image intercepted on the right side of the molten pool tail is wi / 2-|x Pi -x Ei |、Height value is h i / 2+|y Pi -y Ei |.

[0017] Further, in step ②, the computer processing module performs Otsu's global adaptive threshold segmentation on the welding area image to obtain a global binary image of the molten pool, and detects the maximum circumscribed rectangle of the molten pool contour for the global binary image of the molten pool; in step ③, the left and right side image windows of the molten pool tail or the small window image captured by the small window of the molten pool tail image are linearly stretched, median filtered, Otsu's adaptive threshold segmentation, and Canny edge extraction are performed in sequence to obtain the left and right edge images of the molten pool tail or the molten pool tail edge image.

[0018] Further, in step ⑤, the shape characteristic parameter detection value a of the current molten pool tail is i When filtering, the most recent n≥1 times of the molten pool tail shape characteristic parameter detection value a i The median or average value of is taken as the current sampling value a of the shape characteristic parameter of the tail of the molten pool si .

[0019] The technical solution adopted by the infrared visual sensing detection device for the penetration of the shaking arc base weld of the present invention is: it includes a shaking arc welding torch, a voltage sensor, a shaking arc welding controller, a camera trigger module, an infrared camera and a computer processing module; the shaking arc welding torch includes a motor drive and a feeding mechanism, the motor drive and the feeding mechanism detect the instantaneous shaking position of the welding arc in the groove of the workpiece to be welded relative to the center of the shaking arc welding torch, the shaking arc welding controller controls the rotation of the motor on the motor drive and the feeding mechanism, drives the welding arc at the end of the welding wire to perform periodic reciprocating circular arc trajectory motion around the center of the shaking arc welding torch, and stops at the left and right sides of the groove of the workpiece to be welded respectively, and receives the instantaneous shaking position of the arc The infrared camera focuses on the welding molten pool below the welding arc from the front of the groove of the workpiece to be welded at a bird's-eye view shooting angle; in each arc shaking cycle of the shaking arc base welding process, when the welding arc stops at the left or right side of the groove of the workpiece to be welded and shakes to the right or left within an angle range, the shaking arc welding controller outputs an arc shaking position signal to the camera trigger module, and the voltage sensor inputs the detected arc short-circuit voltage signal to the camera trigger module. When the camera trigger module detects the first arc short-circuit voltage signal that is less than or equal to the reference voltage, the camera trigger module triggers the infrared camera to collect the welding area image at this moment.

[0020] Compared with the prior art, the beneficial effects of the present invention are mainly:

[0021] ① For shaking or swinging arc DC welding, the present invention can synchronously collect images of the welding area at a specific groove position and with little arc light interference based on the arc position signal and the arc short-circuit voltage signal. At the same time, it can also effectively reduce the influence of arc shaking or swinging on the detection of molten pool characteristics, thereby having high penetration sensing detection accuracy.

[0022] ② The present invention can collect at least two frames of welding area images in each arc movement cycle, and when processing the welding area images, it can adjust the position of the small window for capturing the characteristic image of the tail of the molten pool in real time according to the current molten pool shape and welding wire position, thereby improving the penetration sensing detection speed and environmental adaptability.

[0023] ③ The present invention only needs to use one industrial camera, and the device system structure is simple and easy to implement. At the same time, the penetration detection method is highly efficient, has strong anti-interference ability and practicality. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is a principle block diagram of the infrared visual sensing detection device for the penetration of the shaking arc bottom welding seam of the present invention;

[0025] Figure 2 for Figure 1 A magnified schematic diagram of the welding arc shaking trajectory;

[0026] Figure 3 This is a schematic diagram of the global image processing principle during the penetration detection of the shaking arc bottom weld of the present invention;

[0027] Figure 4 It is a schematic diagram of double-window interception during the penetration detection of the shaking arc bottom weld of the present invention;

[0028] Figure 5 For Figure 4 Image processing is performed to obtain a schematic diagram of the characteristic edges on the left and right sides of the tail of the molten pool;

[0029] Figure 6 A schematic diagram of a single window cutout during penetration detection of a shaking arc bottom weld according to the present invention;

[0030] Figure 7 For Figure 6 Perform image processing to obtain a schematic diagram of the characteristic edge of the tail of the molten pool;

[0031] Figure 8 This is a schematic diagram of the tail edge of the molten pool when the base weld is fully penetrated;

[0032] Fig. 9 This is a schematic diagram of the tail edge of the molten pool when the base weld is not fully penetrated;

[0033] Fig.10This is a schematic diagram of the tail edge of the molten pool when the base weld is welded through;

[0034] Fig.11 This is an image processing flow chart of the double-window interception method of the present invention;

[0035] Fig.12 This is an image processing flow chart of the single-window interception method of the present invention;

[0036] Fig.13 This is a diagram of the global image processing process in the dual-window interception method in Embodiment 1-3;

[0037] Fig.14 for Fig.13 The process diagram of extracting the characteristic edge of the tail of the molten pool when the middle base weld is fully penetrated;

[0038] Fig.15 for Fig.13 The process diagram of extracting the characteristic edge of the tail of the molten pool when the middle base weld is not fully penetrated;

[0039] Fig.16 for Fig.13 The process diagram of extracting the characteristic edge of the tail of the molten pool when the middle base weld penetrates;

[0040] Fig.17 for Fig.13 Detection effect diagram;

[0041] Fig.18 This is a diagram of the global image processing process in single window capture in Example 4-6;

[0042] Fig.19 for Fig.18 Detection effect diagram;

[0043] Fig. 20 A comparison chart of the test results of the present invention using double-window and single-window interception tests when the base weld is fully penetrated;

[0044] Fig.21 It is a comparison diagram of the test results of the present invention using double-window and single-window interception tests when the base weld is not fully penetrated;

[0045] Fig. 22 The figure is a comparison chart of the detection results of the double-window and single-window interception detection adopted by the present invention when the base weld is welded through.

[0046] In the figure: 1—bevel of the workpiece to be welded; 2—welding arc; 3—bottom weld; 4—welding wire; 5—wire feeding mechanism; 6—motor drive and power feeding mechanism; 7—conductive rod and nozzle mechanism; 8—welding power supply; 9—voltage sensor; 10—shaking arc welding controller; 11—camera trigger module; 12—infrared camera; 13—infrared filter system; 14—image acquisition card; 15—computer processing module; 16—welding area image; 17-1—small window for capturing the image on the left side of the tail of the molten pool 17-2—small window for capturing the image on the right side of the tail of the molten pool; 17-3—small window for capturing the image on the tail of the molten pool; 18-1—feature edge on the left side of the tail of the molten pool; 18-2—feature edge on the right side of the tail of the molten pool; 19—feature edge on the tail of the molten pool; 20—maximum circumscribed rectangle of the molten pool contour; 21—welding molten pool; 22—fitting edge of the tail of the molten pool when the backing weld is fully penetrated; 23—fitting edge of the tail of the molten pool when the backing weld is not fully penetrated; 24—fitting edge of the tail of the molten pool when the backing weld is welded through;

[0047] u a —Arc short circuit voltage signal; P arc —Arc shaking position signal; V w —Welding speed; θ—Top view shooting angle; d—shooting distance; L—left side of the groove of the workpiece to be welded; R—right side of the groove of the workpiece to be welded; G1G2—center line of the groove of the workpiece to be welded; O i , O i+1 —The projection point of the center of the shaking arc welding torch on the bottom surface of the groove 1 of the workpiece to be welded; A 1i O 1i A 2i —The arc-shaped trajectory of the welding arc relative to the shaking arc welding torch when the welding arc is swung from the left side to the right side of the groove of the workpiece to be welded; A 2i+1 O 1i+1 A 1i+1 —The arc trajectory of the welding arc when it swings from the right side to the left side of the groove of the workpiece to be welded relative to the arc trajectory of the swinging arc welding torch (the lagging arc trajectory A 1i O 1i A 2i The time of one right stop); 1i —Circular arc trajectory A 1i O 1i A 2i The midpoint of 1i+1 —Circular arc trajectory A 2i+ 1O 1i+1 A 1i+1 The midpoint of the arc; α1, α2—the instantaneous shaking angle of the arc after the side wall stay ends; M i 、M i+1 —When the arc instantaneous swing angle is α1, the welding arc 2 is in the circular trajectory A 1i O 1i A 2iand arc trajectory A 2i+1 O 1i+1 A 1i+1 The corresponding position point on i 、N i+1 When the arc momentarily swings at an angle of α2, the welding arc 2 is in the circular trajectory A 1i O 1i A 2i and arc trajectory A 2i+ 1O 1i+1 A 1i+1 The corresponding position point on the image; xOy—global image coordinate system; P i —Wire end point; E i —centre of mass of the largest circumscribed rectangle 20; h i —The height of the maximum circumscribed rectangle 20; w i —The width of the maximum circumscribed rectangle 20; B i —The small window positioning point of the image capture on the left side of the molten pool tail; C i —The positioning point of the small window of the image capture on the right side of the molten pool tail; F i —Capture the small window positioning point of the image at the tail of the molten pool. DETAILED DESCRIPTION

[0048] The present invention is further described in detail below in conjunction with the accompanying drawings and specific implementation methods, but the protection scope of the present invention is not limited to the following embodiments, and all technical methods obtained by equivalent replacement or equivalent transformation forms are within the protection scope of the present invention.

[0049] The infrared visual sensor detection device for detecting the penetration of the bottom welding seam by shaking arc of the present invention is as follows Figure 1 As shown, it includes a shaking arc welding torch, a voltage sensor 9, a shaking arc welding controller 10, a camera trigger module 11, an infrared camera system and a computer image processing system. Among them, the shaking arc welding torch is mainly composed of a motor drive and feeding mechanism 6 and a conductive rod and nozzle mechanism 7. The motor drive and feeding mechanism 6 specifically includes a motor (which can be a common motor or a hollow shaft motor), an arc shaking position detection module, a welding feeding module, etc.; the conductive rod and nozzle mechanism 7 further includes a bent conductive rod, a straight conductive nozzle and a nozzle mechanism, or includes a straight conductive rod, a bent conductive nozzle and a nozzle mechanism; the infrared camera system mainly includes an infrared camera 12 and an infrared filter system 13, and the infrared filter system 13 is coaxially installed on the lens of the infrared camera 12; the computer image processing system mainly includes an image acquisition card 14 and a computer processing module 15.

[0050] The arc shaking position detection module in the motor drive and feeding mechanism 6 is a photoelectric (or electromagnetic) detection device, which is used to detect the instantaneous shaking position of the welding arc 2 in the groove 1 of the workpiece to be welded relative to the center of the shaking arc welding torch. The shaking arc welding controller 10 is connected to the motor and the arc shaking position detection module on the motor drive and feeding mechanism 6 respectively, and the shaking arc welding controller 10 is also connected to the camera trigger module 11. The shaking arc welding controller 10 controls the rotation of the motor on the motor drive and feeding mechanism 6, and receives the arc instantaneous shaking position signal output by the arc shaking position detection module. According to the arc instantaneous shaking position signal output by the arc shaking position detection module, the shaking arc welding controller 10 detects the shaking frequency of the welding arc 2 and the arc shaking position signal P arc and the arc shaking position signal P arc Input camera trigger module 11.

[0051] One output cable of the welding power source 8 is connected to the welding power supply module on the motor drive and power supply mechanism 6, and the other output cable is connected to the workpiece to be welded. Two signal input ends of the voltage sensor 9 are connected in parallel to the two output cables of the welding power source 8, and the signal output end of the voltage sensor 9 is connected to the camera trigger module 11. The signal output end of the camera trigger module 11 is connected to the image capture trigger signal input end of the infrared camera 12, and the signal output end of the infrared camera 12 is connected to the input end of the image acquisition card 14 through a video transmission cable, and the output end of the image acquisition card 14 is connected to the computer processing module 15.

[0052] Before starting welding, first adjust the position of the shaking arc welding torch in the width direction of the groove 1 of the workpiece to be welded (i.e., perpendicular to the welding speed V w direction), so that the center of the shaking arc welding torch is projected at point O on the bottom surface of the groove 1 of the workpiece to be welded i (or O i+1 ) is located on the center line G1G2 of the bottom surface of the groove of the workpiece to be welded (such as Figure 2 As shown), the arc shaking parameters (including shaking frequency, shaking angle, left dwell time and right dwell time, etc.), arc energy parameters (including welding current, arc voltage, etc.) and welding speed V are set by shaking the arc welding controller 10, the welding power source 8 and the shaking arc welding torch dragging system. w Then the welding wire 4 sent out by the wire feeding mechanism 5 passes through the shaking arc welding torch and extends obliquely into the groove 1 of the workpiece to be welded, and the shaking arc welding controller 10 controls the motor drive and the motor rotation on the feeding mechanism 6, so that the welding wire 4 points to the front or rear of the welding when the motor stops, that is, the front positioning of the spot welding is realized during the arc shaking (at this time point O 1i Or click O 1i+1After the welding arc 2 is ignited, the shaking arc welding controller 10 controls the motor drive and the motor on the feeding mechanism 6 to rotate, driving the welding arc 2 at the end of the welding wire 4 to perform periodic reciprocating circular arc trajectory motion around the center of the shaking arc welding torch, and stays at the left and right sides of the groove 1 of the workpiece to be welded for a certain period of time (generally tens to hundreds of milliseconds).

[0053] like Figure 2 As shown, when the projection of the center of the shaking arc welding torch on the bottom surface of the groove 1 of the workpiece to be welded is located at point O i When the welding arc 2 stops staying on the left side of the groove 1 of the workpiece to be welded, it starts to swing to the right side of the groove 1 of the workpiece to be welded. Compared with the swinging arc welding torch, the motion trajectory of the welding arc 2 is a circular arc trajectory A. 1i O 1i A 2i , accordingly, the welding arc 2 starts from point A 1i Through the arc track A 1i O 1i A 2i The midpoint O 1i Move to point A 2i Then, when the welding arc 2 stops at the right side of the groove 1 of the workpiece to be welded, the projection of the center of the arc welding torch on the bottom surface of the groove 1 of the workpiece to be welded is moved to point O along the welding direction. i+1 Then, the welding arc 2 starts from point A. 2i+1 Through the arc track A 2i+1 O 1i+ 1A 1i+1 The midpoint O 1i+1 Move to point A 1i+1 , and will stop at the left side of the groove 1 of the workpiece to be welded. Then, the welding arc 2 will start to swing to the right side of the groove of the workpiece to be welded again, and repeat the above-mentioned periodic reciprocating circular arc trajectory motion relative to the swinging arc welding torch until the welding stops.

[0054] like Figure 1 and Figure 3As shown, in order to quickly collect a clear welding area image 16, the present invention preferably uses a CMOS-type infrared camera 12, and the infrared camera 12 is fixedly connected to the front of the welding of the shaking arc welding torch, so that the infrared camera 12 is focused on the welding pool 21 below the welding arc 2 at a 30° to 45° downward shooting angle θ from the front of the groove 1 of the workpiece to be welded, and the shooting distance d from the infrared filter system 13 to the upper surface of the welding pool 21 is 200 to 300 mm. The infrared filter system 13 used includes a narrow-band filter with a central wavelength of 808 to 970 nm, a 5 to 30% transmittance dimming filter and a UV filter. Among them, the narrow-band filter is mainly used to filter out light radiation interference such as arc light, smoke, and splashes; the dimming filter is mainly used to adjust the arc and molten pool radiation light intensity entering the infrared camera 12, so that a clear infrared image of the welding area can be collected; the UV filter mainly plays a protective role for the narrow-band filter and the dimming filter.

[0055] like Figure 1 and Figure 2 As shown, during the shaking arc rooting welding process, in each arc shaking cycle, when the welding arc 2 stops at the left side of the groove 1 of the workpiece to be welded and shakes to the right within the angle range of α1 to α2 ( Figure 2 The welding arc 2 is relative to the shaking arc welding torch from point M i Move to point N i During this period, α1<α2), or when the welding arc 2 stops at the right side of the groove 1 of the workpiece to be welded and swings to the left within the angle range of α1 to α2 ( Figure 2 The welding arc 2 is relative to the shaking arc welding torch from point M i+1 Move to point N i+1 During this period, α1<α2), the shaking arc welding controller 10 outputs the arc shaking position signal P to the camera trigger module 11. arc During this period, the voltage sensor 9 simultaneously detects the arc short-circuit voltage signal u a Input camera trigger module 11. The instantaneous arc shaking angle α1 after the side wall stay is preferably in the range of 1 / 4 to 1 / 2 of the arc shaking angle setting value, and the instantaneous arc shaking angle α2 is preferably in the range of 1 / 2 to 3 / 4 of the arc shaking angle setting value.

[0056] The oscillating arc welding controller 10 outputs an arc oscillation position signal P arc During this period, when the camera trigger module 11 detects the first voltage that is less than or equal to the reference voltage V REF (Reference voltage V REF The arc short-circuit voltage signal u is preferably in the range of 0 to 10 V. aWhen the camera trigger module 11 outputs a camera shooting trigger signal, the infrared camera 12 is immediately triggered to collect the welding area image 16 at this moment, so as to realize the collection of the welding area image 16 in which the arc shaking position is synchronized with the short-circuit moment of the welding arc 2, thereby reducing the interference of the arc light on the extraction of the characteristic edge of the tail of the base welding molten pool.

[0057] The oscillating arc welding controller 10 outputs an arc oscillation position signal P arc During this period, if the camera trigger module 11 does not detect a voltage less than or equal to the reference voltage V REF Arc short-circuit voltage signal u a When the camera trigger module 11 outputs the arc shaking position signal P when the arc welding controller 10 shakes arc At the last moment (i.e., when the welding arc 2 finishes staying on the left side of the groove 1 of the workpiece to be welded and swings to the right by an angle of α2, or when the welding arc 2 finishes staying on the right side of the groove 1 of the workpiece to be welded and swings to the left by an angle of α2), the camera trigger module 11 outputs a camera shooting trigger signal, which immediately triggers the infrared camera 2 to capture the welding area image 16 at this moment.

[0058] See also Figures 1 to 12 The specific steps of the infrared visual sensing detection method for the penetration of the shaking arc bottom weld of the present invention are as follows:

[0059] ① Image acquisition. During the shaking arc root welding process, the infrared camera system and the shaking arc welding torch move at a welding speed V w Move forward to the groove 1 of the workpiece to be welded, and the camera trigger module 11 is triggered according to the arc position signal P arc And arc short circuit voltage signal u a Output camera shooting trigger signal, trigger infrared camera 12 to collect the current Figure 3 As shown in the welding area image 16 , the image acquisition card 14 sends the welding area image 16 acquired by the infrared camera 12 to the computer processing module 15 .

[0060] ② Global image processing. First, the computer processing module 15 performs Otsu's global adaptive threshold segmentation on the received welding area image 16 to obtain a global binary image of the molten pool. Then, for the global binary image of the molten pool, the maximum circumscribed rectangle 20 (such as Figure 3 As shown), extract the centroid E of the maximum circumscribed rectangle 20 i Coordinate value (x Ei ,y Ei ), height value h i and width w i Next, for the global binary image of the molten pool, the tail image of the molten pool is intercepted. The method for intercepting the tail image of the molten pool can adopt a double-window interception method or a single-window interception method, specifically:

[0061] When the double window interception method is used: firstly, the end point P of the welding wire is extracted from the inside of the maximum circumscribed rectangle 20. i Coordinate value (x Pi ,y Pi ). Wire end point P i The specific method of extracting coordinates is to use the maximum circumscribed rectangle 20 centroid E on the global binary image of the melt pool. i (x Ei ,y Ei ) is centered in an area of ​​80 pixels × 80 pixels, and the pixel point with a gray value of 0 and the smallest ordinate value is searched, and the coordinate of the pixel point is used as the end point P of the welding wire. i Coordinate (x Pi ,y Pi ). Then, according to the maximum circumscribed rectangle 20 centroid E i Coordinate value, height value h i and width w i Adjust the positioning point B in the small image window 17-1 on the left side of the tail of the molten pool. i The coordinate value of the small image window 17-2 on the right side of the molten pool tail, the positioning point C i Coordinate values; according to the centroid E of the maximum circumscribed rectangle 20 i Coordinate value, height value h i and width w i , and the end point P of the welding wire i The change of coordinate values ​​adjusts the height and width of the small image window 17-1 on the left side of the molten pool tail and the small image window 17-2 on the right side of the molten pool tail, such as Figure 4 shown.

[0062] The method for determining the left image window 17-1 of the tail of the molten pool and the right image window 17-2 of the tail of the molten pool is specifically to determine the center of gravity E of the maximum circumscribed rectangle 20 according to the center of gravity E of the maximum circumscribed rectangle 20. i Coordinate value, height value h i and width w i , determine the image capture window 17-1 positioning point B on the left side of the molten pool tail i The coordinate value of the image capture window 17-2 on the right side of the tail of the molten pool and the positioning point C i The horizontal coordinate value is set, and the small image window 17-2 on the right side of the tail of the molten pool locates point C i The vertical coordinate value is the same as the positioning point B in the small image window 17-1 on the left side of the molten pool tail i The vertical coordinate values ​​are equal. i The horizontal axis value x Pi Less than or equal to the center of gravity E of the maximum circumscribed rectangle 20 i The horizontal axis value xEi When (i.e. x Pi ≤x Ei ), the width of the small image window 17-1 on the left side of the molten pool tail is w i / 2-|x Pi -x Ei |、Height value is h i / 2+|y Pi -y Ei |, the width of the small window 17-2 of the image capture on the right side of the molten pool tail is w i / 2+|x Pi -x Ei |、Height value is h i / 2-|y Pi -y Ei |; When the end point P of the welding wire i The horizontal axis value x Pi Greater than the maximum circumscribed rectangle 20 centroid E i The horizontal axis value x Ei When (i.e. x Pi >x Ei ), the width of the small window 17-1 for capturing the image on the left side of the tail of the molten pool is w i / 2+|x Pi -x Ei |、Height value is h i / 2-|y Pi -y Ei |, the width of the small window 17-2 of the image capture on the right side of the molten pool tail is w i / 2-|x Pi -x Ei |、Height value is h i / 2+|y Pi -y Ei |. Anchor point B i Coordinate value, positioning point C i The horizontal coordinate value is based on the center of gravity E of the maximum circumscribed rectangle 20 i Coordinate value, height value h i and width w i It will be adjusted dynamically according to the maximum circumscribed rectangle 20, and the adjustment range is fixed under the same test conditions.

[0063] When the single window interception method is used: according to the maximum circumscribed rectangle 20 centroid E i Coordinate value, height value h i and width w i Changes, adjust the positioning point F of the small window 17-3 of the tail image of the molten pool i The coordinate value of the maximum circumscribed rectangle 20 i and width w i, adjust the height and width of the small window 17-3 for capturing the tail image of the molten pool, such as Figure 6 shown.

[0064] ③ Image processing of the tail of the molten pool in a small window.

[0065] For the small window 17-1 of the left side image of the tail of the molten pool and the small window 17-2 of the right side image of the tail of the molten pool obtained by the double window interception method, first, linear stretching, median filtering, Otsu's method adaptive threshold segmentation, and Canny edge extraction are performed in sequence to obtain the left side edge image of the tail of the molten pool and the right side edge image of the tail of the molten pool; then, in the left side edge image of the tail of the molten pool, the highest point of the left side edge of the tail of the molten pool and the maximum point of the edge curvature searched from the highest point to the left are extracted, and the edge located between the highest point and the maximum point of the curvature in the left side edge image of the tail of the molten pool is intercepted as the left side characteristic edge 18-1 of the tail of the molten pool; at the same time, in the right side edge image of the tail of the molten pool, the highest point of the right side edge of the tail of the molten pool and the maximum point of the edge curvature searched from the highest point to the right are extracted, and the edge located between the highest point and the maximum point of the curvature in the right side edge image of the tail of the molten pool is intercepted as the right side characteristic edge 18-2 of the tail of the molten pool, as shown in FIG. Figure 5 shown.

[0066] For the small window 17-3 of the molten pool tail image captured by a single window, first, linear stretching, median filtering, Otsu's method adaptive threshold segmentation, and Canny edge extraction are performed in sequence to obtain the molten pool tail edge image; then, in the part above half of the height of the molten pool tail edge image, the first extreme point of the vertical coordinate of the molten pool tail edge is searched from the left side of the image to the right, and the maximum point of edge curvature is searched from the extreme point to the left, and the edge between the extreme point and the maximum point of edge curvature is intercepted as the left part of the characteristic edge of the molten pool tail; at the same time, the first extreme point of the vertical coordinate of the molten pool tail edge is searched from the right side of the image to the left, and the maximum point of edge curvature is searched from the extreme point to the right, and the edge between the extreme point and the maximum point of edge curvature is intercepted as the right part of the characteristic edge of the molten pool tail; the intercepted left part of the characteristic edge of the molten pool tail and the right part of the characteristic edge of the molten pool tail are collectively used as the characteristic edge of the molten pool tail 19, as shown in FIG. Figure 7 shown.

[0067] ④ Extraction of penetration status of base weld

[0068] For the double-window interception method, the obtained left characteristic edge 18-1 of the tail of the molten pool and the right characteristic edge 18-2 of the tail of the molten pool are first mapped to the global image coordinate system xOy at the same time; then, in the global image coordinate system xOy, the quadratic curve fitting method is used to perform quadratic curve fitting calculation on the extracted left characteristic edge 18-1 of the tail of the molten pool and the right characteristic edge 18-2 of the tail of the molten pool to obtain the fitting edge 22 (or 23 or 24) of the tail of the molten pool, as shown in FIG. Figure 8 , Fig. 9 , Fig.10 As shown, the curve equation of the fitting edge 22 (or 23 or 24) of the tail of the molten pool in the global image coordinate system xOy is obtained; then, the computer processing module 15 extracts the quadratic term coefficient from the quadratic curve equation of the fitting edge 22 (or 23 or 24) of the tail of the molten pool, and uses the quadratic term coefficient as the current molten pool tail shape feature parameter detection value a i ; Finally, the shape characteristic parameter detection value a of the tail of the molten pool for the most recent n times is used i The median or average value of is taken as the current sampling value a of the shape characteristic parameter of the tail of the molten pool si (n≥1), if a si ≤a min The current weld is in the weld-through state. min <a si <a max The current weld is in the penetration state. If a si ≥a max The current weld is in an incomplete penetration state. min is the criterion for the penetration of the swing arc backing weld, a max It is the criterion for judging whether the shaking arc base weld is not fully penetrated.

[0069] For the single window interception method, the obtained characteristic edge 19 of the tail of the molten pool is first mapped to the global image coordinate system xOy; then, in the global image coordinate system xOy, a quadratic curve fitting method is used to perform a quadratic curve fitting calculation on the extracted characteristic edge 19 of the tail of the molten pool to obtain the fitting edge 22 (or 23 or 24) of the tail of the molten pool, and the curve equation of the fitting edge 22 (or 23 or 24) of the tail of the molten pool in the global image coordinate system xOy is obtained; then, the computer processing module 15 extracts the quadratic term coefficient from the quadratic curve equation of the fitting edge 22 (or 23 or 24) of the tail of the molten pool, and uses the quadratic term coefficient as the current molten pool tail shape feature parameter detection value a i Similarly, the final value a of the shape characteristic parameter of the tail of the molten pool is taken as the last n times i The median or average value of is taken as the current sampling value a of the shape characteristic parameter of the tail of the molten pool si (n≥1), if a si ≤a min The current weld is in the weld-through state. min <a si <a max The current weld is in the penetration state. If a si ≥a max The current weld is in an incomplete penetration state. min is the criterion for the penetration of the swing arc backing weld, a maxIt is the criterion for judging whether the shaking arc base weld is not fully penetrated.

[0070] For example, when the groove 1 of the workpiece to be welded is a U-shaped groove with a groove gap of 12mm, a root gap of 2mm, and a blunt edge thickness of 2mm, the shaking frequency is 1.0-3.5Hz, the shaking angle is 40-80°, the left dwell time and the right dwell time are 40-100ms, and the welding speed V w When the arc current is 15.5cm / min, the arc voltage is 150-165A, and the arc voltage is 19-20.5V, the criterion for welding penetration of the swing arc backing weld is a. min The preferred value is -0.009, and the criterion for incomplete penetration of the shaking arc backing weld is a max The preferred value is -0.007.

[0071] In one arc shaking cycle, the welding arc 2 experiences one shaking from the left side to the right side of the groove 1 of the workpiece to be welded and one shaking from the right side to the left side of the groove 1 of the workpiece to be welded, so that two detections of the penetration of the shaking arc backing weld can be realized, thereby improving the real-time performance of the penetration detection of the shaking arc backing weld and leaving sufficient adjustment time for the subsequent penetration control of the shaking arc backing weld 3. In addition, by intercepting the image of the tail of the molten pool when the welding arc 2 is short-circuited within a specific instantaneous shaking angle range through a small window for processing, the interference of arc shaking and its arc light can be avoided, and the detection accuracy of the penetration of the backing weld can be improved.

[0072] In the infrared visual sensing detection method for the vibration arc welding penetration of the present invention, image processing is a key step. The image processing detection processes for the vibration arc welding penetration state of the two methods of double window interception and single window interception are respectively as follows: Fig.11 and Fig.12 shown.

[0073] Fig.11 The figure shows a flow chart of the double window interception method. In the first step of global image processing, the computer processing module 15 performs global adaptive threshold segmentation of the received welding area image 16 by the Otsu method, and extracts the centroid E of the maximum circumscribed rectangle 20 of the molten pool contour on the global binary image of the molten pool obtained by the threshold segmentation. i Coordinate value, height value h i and width w i , and the end point P of the welding wire i Coordinate value; according to the maximum circumscribed rectangle 20 of the molten pool contour and the end point P of the welding wire iThe position of the small window 17-1 for capturing the left image of the tail of the molten pool and the small window 17-2 for capturing the right image of the tail of the molten pool are adaptively determined. In the second step of processing the image of the tail of the molten pool in the small window, linear stretching, median filtering, Otsu's method adaptive threshold segmentation, and Canny edge extraction are performed on the captured small window image of the left side of the tail of the molten pool and the small window image of the right side of the tail of the molten pool, respectively, to obtain the left edge image of the tail of the molten pool and the right edge image of the tail of the molten pool, and then the highest point of the left edge of the tail of the molten pool and the highest point of the right edge of the tail of the molten pool are extracted respectively, and the maximum point of the edge curvature is searched to the left and right from the two highest points of the edge of the tail of the molten pool, respectively, and the edge between the highest point of the left edge of the tail of the molten pool to the maximum point of the curvature of the left edge is extracted as the left characteristic edge 18-1 of the tail of the molten pool, and the edge between the highest point of the right edge of the tail of the molten pool to the maximum point of the curvature of the right edge is extracted as the right characteristic edge 18-2 of the tail of the molten pool. In the third step, when extracting the weld penetration state, for the extracted left characteristic edge 18-1 of the molten pool tail and the right characteristic edge 18-2 of the molten pool tail, the molten pool tail fitting edge 22 (or 23 or 24) and its curve equation are obtained by quadratic curve fitting, and the quadratic term coefficient of the quadratic curve equation is extracted as the current molten pool tail shape characteristic parameter detection value a i , and the most recent n times the shape characteristic parameter detection value a of the molten pool tail i The median or average value of is taken as the current sampling value a of the shape characteristic parameter of the tail of the molten pool si , and finally detect the current sampling value a si The criterion for incomplete penetration of the swing arc backing weld is a max and the criterion for penetration of the oscillating arc backing weld a min The size relationship is calculated and the bottom weld penetration status information is output for the subsequent shaking arc bottom weld penetration control.

[0074] Fig.12 The flowchart of the single window interception method is shown. In the first step of global image processing, the computer processing module 15 performs global adaptive threshold segmentation of the received welding area image 16 by the Otsu method, and extracts the center of gravity E of the maximum circumscribed rectangle 20 of the molten pool contour on the global binary image of the molten pool obtained by the threshold segmentation. i Coordinate value, height value h i and width w i; According to the change of the maximum circumscribed rectangle 20 feature of the molten pool contour, the position of the small window 17-3 for intercepting the molten pool tail image is adaptively determined. In the second step, when processing the small window molten pool tail image, the intercepted small window image of the molten pool tail is subjected to linear stretching, median filtering, Otsu's method adaptive threshold segmentation, and Canny edge extraction in sequence to obtain the molten pool tail edge image, and then the first extreme point of the ordinate of the molten pool tail edge is searched from the left side of the image to the right and from the right side of the image to the left in more than half of the height of the molten pool tail edge image, and the maximum point of edge curvature is searched from these two extreme points to the left and right in the molten pool tail edge image, and at the same time, the molten pool tail edge between the extreme point on the left side of the molten pool tail and the maximum point of edge curvature on the left side, and the molten pool tail edge between the extreme point on the right side of the molten pool tail and the maximum point of edge curvature on the right side are extracted as the characteristic edge 19 of the molten pool tail. In the third step, when extracting the weld penetration state, for the extracted characteristic edge 19 of the molten pool tail, the fitting edge 22 (or 23 or 24) of the molten pool tail and its curve equation are obtained by quadratic curve fitting, and the quadratic term coefficient of the quadratic curve equation is extracted as the current molten pool tail shape characteristic parameter detection value a i , and the most recent n times the shape characteristic parameter detection value a of the molten pool tail i The median or average value of is taken as the current sampling value a of the shape characteristic parameter of the tail of the molten pool si , and finally detect the current sampling value a si The criterion for incomplete penetration of the swing arc backing weld is a max and the criterion for penetration of the oscillating arc backing weld a min The size relationship is calculated and the bottom weld penetration status information is output for the subsequent shaking arc bottom weld penetration control.

[0075] The following are six embodiments of the infrared visual sensing detection method of the present invention. The specific test conditions include: the infrared camera 12 is a CMOS infrared camera, the lens aperture is f16, the exposure time is 0.3ms, the top view shooting angle θ is 30°; the shooting distance d is 250mm; the infrared filter system 13 includes a narrow band filter with a central wavelength of 808nm, a dimming filter with a transmittance of 30%, and a protective UV filter. The material of the workpiece to be welded is Q370 steel, the groove 1 of the workpiece to be welded is a U-shaped groove with a groove gap of 12mm, a root gap of 2mm, and a blunt edge thickness of 2mm. The welding current is 150-165A, the arc voltage is 19-20.5V, and the welding speed V w The arc shaking parameters are as follows: the shaking frequency is 1.0 to 3.5 Hz, the shaking angle is 40° to 80°, and the left dwell time and the right dwell time are 40 to 100 ms.

[0076] Example 1

[0077] Fig.13 It is the global image processing result in the double window interception detection method of the swing arc bottom weld penetration of the present invention. Among them, Fig.13 a1 is a frame of welding area image 16 captured by infrared camera 12 when the base weld is welded through. The shaking frequency used is 2.0 Hz, the shaking angle is 60°, the left dwell time and the right dwell time are both 70 ms, the average welding current is 163 A, the average arc voltage is 19.8 V, and the reference voltage V REF The voltage is 7V, and the size (width×height) of the acquired welding area image 16 is 512 pixels×512 pixels. At this moment, the welding arc 2 ends and stays on the right side of the groove 1 of the workpiece to be welded and swings 29° to the left. Fig.13 a1, after global adaptive threshold segmentation using Otsu method, the global binary image of the melt pool is obtained as follows Fig.13 As shown in a2, the white area represents the molten pool area, and the rod-shaped black area in the molten pool area is the welding wire. Fig.13 The maximum circumscribed rectangle 20 of the molten pool contour detected on a2 is as follows Fig.13 a3. In the global coordinate system xOy, the center of gravity E of the maximum circumscribed rectangle 20 is i The coordinates are (238, 268) and the height value is h i is 254, width value w i is 212; in the maximum circumscribed rectangle 20 centroid E i The welding wire end point P is searched within the 80 pixels × 80 pixels area centered at i The coordinates are (237, 245). Accordingly, the double-window interception method is used to locate the left image window 17-1 of the tail of the molten pool and the right image window 17-2 of the tail of the molten pool. Fig.13 As shown in a4.

[0078] Fig.14 Shown for Fig.13 a1 is an example of the feature edge extraction process of the tail of the molten pool when the base weld is welded through. Fig.14 a1) The image of the right side of the molten pool tail captured by the small window 17-2 (see Fig.14 b1), after linear stretching and median filtering denoising, the obtained image is as follows Fig.14 a2 and Fig.14 As shown in b2. Fig.14 a2 and Fig.14After b2 is segmented by the global adaptive threshold of the Otsu method, the binary image of the small window on the left side of the tail of the molten pool and the binary image of the small window on the right side of the tail of the molten pool are obtained as follows: Fig.14 a3 and Fig.14 b3. Then, the Canny edge detection operator is used to Fig.14 a3 and Fig.14 b3 performs edge extraction, and the obtained left edge image of the molten pool tail and the right edge image of the molten pool tail are respectively as follows Fig.14 a4 and Fig.14 As shown in b4. Fig.14 a4 and Fig.14 In b4, the extracted left characteristic edge 18-1 of the molten pool tail and the right characteristic edge 18-2 of the molten pool tail are as shown in FIG. Fig.14 a5 and Fig.14 As shown in b5.

[0079] like Fig.17 As shown, Fig.14 a5 and Fig.14 The characteristic edge image of the tail of the molten pool shown in b5 is simultaneously mapped to the global coordinate system xOy, and a quadratic curve is used for fitting calculation to obtain the fitted edge image of the tail of the molten pool as shown in Fig.17 As shown in a2, the coefficient of the quadratic term of the curve equation is -0.00767, which is the current detection value of the characteristic parameter of the tail shape of the molten pool a. i is -0.00767. In addition, the position of the edge of the tail of the molten pool after fitting in the original welding area image is as follows: Fig.17 As shown in a3, the corresponding cross-section formation photo of the bottom welding head is as follows Fig.17 As shown in a4. Fig.17 It can be seen in a3 that the shape and position of the edge of the tail of the molten pool after fitting reflect the state of the tail of the molten pool in the original welding area image.

[0080] Example 2

[0081] Fig.13 b1 shows a welding area image 16 captured by the infrared camera 12 when the base weld is not fully penetrated. The shaking frequency used is 3.5 Hz, the shaking angle is 80°, the left dwell time and the right dwell time are both 100 ms, the average welding current is 161 A, the average arc voltage is 19.7 V, and the reference voltage V REF The voltage is 6V, and the size of the acquired welding area image 16 is also 512 pixels × 512 pixels. At this moment, the welding arc 2 ends and stays on the right side of the groove 1 of the workpiece to be welded and swings 28° to the left. Fig.13 b1, after global adaptive threshold segmentation using Otsu method, the global binary image of the melt pool is obtained as follows Fig.13 b2. Then, Fig.13The maximum circumscribed rectangle 20 of the molten pool contour detected on b2 is as follows Fig.13 b3. In the global coordinate system xOy, the center of gravity E of the maximum circumscribed rectangle 20 is i The coordinates are (241, 291) and the height value is h i is 236, width value w i is 220; in the maximum circumscribed rectangle 20 centroid E i The welding wire end point P is searched within the 80 pixels × 80 pixels area centered at i The coordinates of are (271, 267). Accordingly, the left image window 17-1 of the molten pool tail and the right image window 17-2 of the molten pool tail located by the double-window interception detection method of the present invention are as follows: Fig.13 As shown in b4.

[0082] Fig.15 For Fig.13 b1 is an example of the feature edge extraction process of the tail of the molten pool when the base weld is not fully penetrated. Fig.15 a1) The image of the right side of the molten pool tail captured by the small window 17-2 (see Fig.15 b1), after linear stretching and median filtering denoising, the obtained image is as follows Fig.15 a2 and Fig.15 As shown in b2. Fig.15 a2 and Fig.15 After b2 is segmented by the global adaptive threshold of the Otsu method, the binary image of the small window on the left side of the tail of the molten pool and the binary image of the small window on the right side of the tail of the molten pool are obtained as follows: Fig.15 a3 and Fig.15 b3. Next, the Canny edge detection operator is used to Fig.15 a3 and Fig.15 b3 performs edge extraction, and the obtained left edge image of the molten pool tail and the right edge image of the molten pool tail are respectively as follows Fig.15 a4 and Fig.15 As shown in b4, the extracted left characteristic edge 18-1 of the tail of the molten pool and the right characteristic edge 18-2 of the tail of the molten pool are as shown in Fig.15 a5 and Fig.15 As shown in b5.

[0083] like Fig.17 As shown, Fig.15 a5 and Fig.15 The characteristic edge image of the tail of the molten pool shown in b5 is simultaneously mapped to the global coordinate system xOy, and a quadratic curve is used for fitting calculation to obtain the fitted edge image of the tail of the molten pool as shown in Fig.17As shown in b2, the coefficient of the quadratic term of the curve equation is -0.00638, which is the detection value of the current shape characteristic parameter of the molten pool tail a. i is -0.00638. In addition, the position of the edge of the tail of the molten pool after fitting in the original welding area image is as follows: Fig.17 As shown in b3, the corresponding cross-section formation photo of the bottom welding head is as follows Fig.17 As shown in b4. Fig.17 It can be seen from b3 that the shape and position of the edge of the tail of the molten pool after fitting reflect the state of the tail of the molten pool in the original welding area image.

[0084] Example 3

[0085] Fig.13 c1 shows a welding area image 16 captured by the infrared camera 12 when the base weld is welded through. The shaking frequency used is 1.0 Hz, the shaking angle is 40°, the left dwell time and the right dwell time are both 40 ms, the average welding current is 162 A, the average arc voltage is 19.8 V, and the reference voltage V REF The voltage is 10V, and the size of the acquired welding area image 16 is also 512 pixels × 512 pixels. At this moment, the welding arc 2 ends and stays on the left side of the groove 1 of the workpiece to be welded and swings 26° to the right. Fig.13 c1, after global adaptive threshold segmentation using Otsu method, the global binary image of the melt pool is obtained as follows Fig.13 c2. Then, Fig.13 The maximum circumscribed rectangle 20 of the melt pool contour detected on c2 is as follows Fig.13 c3. In the global coordinate system xOy, the center of gravity E of the maximum circumscribed rectangle 20 is i The coordinates are (254, 260) and the height value is h i is 284, width value w i is 206; in the maximum circumscribed rectangle 20 centroid E i The welding wire end point P is searched within the 80 pixels × 80 pixels area centered at i The coordinates of are (228, 232). Accordingly, the left image window 17-1 of the tail of the molten pool and the right image window 17-2 of the tail of the molten pool located by the double window interception method are as follows: Fig.13 As shown in c4.

[0086] Fig.16 For Fig.13 c1 is an example of the feature edge extraction process of the tail of the molten pool when the base weld is welded through. The left small window image of the tail of the molten pool captured by the small window 17-1 (see Fig.16 a1) The image of the right side of the tail of the molten pool is captured by the small window 17-2 (see Fig.16 b1), after linear stretching and median filtering denoising, the obtained image is as follows Fig.16 a2 and Fig.16 As shown in b2. Fig.16 a2 and Fig.16 After b2 is segmented by the global adaptive threshold of the Otsu method, the binary image of the small window on the left side of the tail of the molten pool and the binary image of the small window on the right side of the tail of the molten pool are obtained as follows: Fig.16 a3 and Fig.16 b3. Next, the Canny edge detection operator is used to Fig.16 a3 and Fig.16 b3 performs edge extraction, and the obtained left edge image of the molten pool tail and the right edge image of the molten pool tail are respectively as follows Fig.16 a4 and Fig.16 As shown in b4. Fig.16 In Fig. 1664, the extracted left characteristic edge 18-1 of the tail of the molten pool and the right characteristic edge 18-2 of the tail of the molten pool are as shown in Fig. Fig.16 a5 and Fig.16 As shown in b5.

[0087] like Fig.17 As shown, Fig.16 a5 and Fig.16 The characteristic edge image of the tail of the molten pool shown in b5 is simultaneously mapped to the global coordinate system xOy, and a quadratic curve is used for fitting calculation to obtain the fitted edge image of the tail of the molten pool as shown in Fig.17 As shown in c2, the coefficient of the quadratic term of the curve equation is -0.01084, which is the detection value of the current shape characteristic parameter of the molten pool tail a i is -0.01084. In addition, the position of the tail edge of the fitted molten pool in the original welding area image is as follows: Fig.17 As shown in c3, the corresponding cross-section formation photo of the bottom welding head is as follows Fig.17 As shown in c4. Fig.17 It can be seen from c3 that the shape and position of the edge of the tail of the molten pool after fitting reflect the state of the tail of the molten pool in the original welding area image.

[0088] Example 4

[0089] Fig.18 The figure shows the global image processing result using the single window interception detection method. Fig.18 a1 shows a welding area image 16 captured by the infrared camera 12 when the base weld is fully penetrated. The shaking frequency used is 2.5 Hz, the shaking angle is 55°, the left dwell time and the right dwell time are both 60 ms, the average welding current is 163 A, the average arc voltage is 19.8 V, and the reference voltage V REFThe voltage is 1V, and the size (width×height) of the acquired welding area image 16 is 512 pixels×512 pixels. At this moment, the welding arc 2 ends and stays on the left side of the groove 1 of the workpiece to be welded and swings 20° to the right. Fig.18 a1, after global adaptive threshold segmentation using Otsu method, the global binary image of the melt pool is obtained as follows Fig.18 As shown in a2, the white area represents the molten pool area, and the rod-shaped black area in the molten pool area is the welding wire. Fig.18 The maximum circumscribed rectangle 20 of the molten pool contour detected on a2 is as follows Fig.18 a3. In the global coordinate system xOy, the center of gravity E of the maximum circumscribed rectangle 20 is i The coordinates are (238, 258) and the height value is h i is 253, width value w i is 214. Accordingly, the small window 17-3 for capturing the image of the tail of the molten pool positioned by the single window capture detection method of the present invention is as follows: Fig.18 As shown in a4.

[0090] The image of the tail of the molten pool captured by the small window 17-3 (see Fig.19 a1), after linear stretching and median filtering denoising, the obtained image is as follows Fig.19 As shown in a2. Fig.19 After a2 performs global adaptive threshold segmentation using the Otsu method, a small window binary image of the tail of the molten pool is obtained. Fig.19 Then, the Canny edge detection operator is used to Fig.19 a3 performs edge extraction, and the obtained edge image of the molten pool tail is as follows Fig.19 As shown in a4, the extracted characteristic edge of the molten pool tail is as follows Fig.19 As shown in a5. Fig.19 The characteristic edge image of the tail of the molten pool shown in a5 is mapped to the global coordinate system xOy, and a quadratic curve is used for fitting calculation to obtain the fitted edge image of the tail of the molten pool as shown in Fig.19 As shown in a6, the coefficient of the quadratic term of the curve equation is -0.00794, which is the current detection value of the shape characteristic parameter of the tail of the molten pool a i is -0.00794. In addition, the position of the edge of the tail of the molten pool after fitting in the original welding area image is as follows: Fig.19 As shown in a7, it can be seen that the shape and position of the edge of the molten pool tail after fitting also reflect the state of the molten pool tail in the original welding area image.

[0091] Example 5

[0092] Fig.18b1 shows a welding area image 16 captured by the infrared camera 12 when the base weld is not fully penetrated. The shaking frequency used is 3.0 Hz, the shaking angle is 75°, the left dwell time and the right dwell time are both 90 ms, the average welding current is 161 A, the average arc voltage is 19.7 V, and the reference voltage V REF is 0V, and the size (width×height) of the acquired welding area image 16 is 512 pixels×512 pixels. At this moment, the welding arc 2 ends and stays on the left side of the groove 1 of the workpiece to be welded and swings 42° to the right. Fig.18 b1, after global adaptive threshold segmentation using Otsu method, the global binary image of the melt pool is obtained as follows Fig.18 b2. Then, Fig.18 The maximum circumscribed rectangle 20 of the molten pool contour detected on b2 is as follows Fig.18 b3. In the global coordinate system xOy, the center of gravity E of the maximum circumscribed rectangle 20 is i The coordinates are (242, 286) and the height value is h i The width value is 235. i is 224. Accordingly, the small window 17-3 of the tail image of the molten pool located by the single window interception detection method of the present invention is as follows: Fig.18 As shown in b4.

[0093] The image of the tail of the molten pool captured by the small window 17-3 (see Fig.19 b1), after linear stretching and median filtering denoising, the obtained image is as follows Fig.19 As shown in b2. Fig.19 After b2 is segmented by the global adaptive threshold of the Otsu method, the binary image of the small window at the tail of the molten pool is obtained as follows: Fig.19 b3. Next, the Canny edge detection operator is used to Fig.19 b3 performs edge extraction, and the obtained edge image of the molten pool tail is as follows Fig.19 As shown in b4, the extracted characteristic edge of the molten pool tail is as follows Fig.19 As shown in b5. Fig.19 The characteristic edge image of the tail of the molten pool shown in b5 is mapped to the global coordinate system xOy, and a quadratic curve is used for fitting calculation to obtain the fitted edge image of the tail of the molten pool as shown in Fig.19 As shown in b6, the coefficient of the quadratic term of the curve equation is -0.00565, which is the current detection value of the shape characteristic parameter of the tail of the molten pool a. i is -0.00565. In addition, the position of the edge of the tail of the molten pool after fitting in the original welding area image is as follows: Fig.19 As shown in b7, it can be seen that the shape and position of the edge of the molten pool tail after fitting also reflect the state of the molten pool tail in the original welding area image.

[0094] Example 6

[0095] Fig.18 c1 shows a welding area image 16 captured by the infrared camera 12 when the base weld is welded through. The shaking frequency used is 1.5 Hz, the shaking angle is 45°, the left dwell time and the right dwell time are both 50 ms, the average welding current is 162 A, the average arc voltage is 19.8 V, and the reference voltage V REF The voltage is 3V, and the size (width×height) of the acquired welding area image 16 is 512 pixels×512 pixels. At this moment, the welding arc 2 ends and stays on the right side of the groove 1 of the workpiece to be welded and swings 28° to the left. Fig.18 c1, after global adaptive threshold segmentation using Otsu method, the global binary image of the melt pool is obtained as follows Fig.18 c2. Then, Fig.18 The maximum circumscribed rectangle 20 of the melt pool contour detected on c2 is as follows Fig.18 c3. In the global coordinate system xOy, the center of gravity E of the maximum circumscribed rectangle 20 is i The coordinates are (254, 264) and the height value is h i is 278, width value w i is 208. Accordingly, the small window 17-3 of the tail image of the molten pool located by the single window interception detection method of the present invention is as follows: Fig.18 As shown in c4.

[0096] The image of the tail of the molten pool is captured in the small window 17-3 (see Fig.19 c1), after linear stretching and median filtering denoising, the obtained image is as follows Fig.19 c2. Fig.19 After c2 performs global adaptive threshold segmentation using the Otsu method, a small window binary image of the tail of the melt pool is obtained. Fig.19 c3. Next, the Canny edge detection operator is used to Fig.19 C3 performs edge extraction, and the obtained edge image of the molten pool tail is as follows Fig.19 As shown in c4, the extracted characteristic edge of the molten pool tail is as follows Fig.19 As shown in c5. After extracting the characteristic edge of the molten pool tail, Fig.19 The characteristic edge image of the tail of the molten pool shown in c5 is mapped to the global coordinate system xOy, and a quadratic curve is used for fitting calculation to obtain the fitted edge image of the tail of the molten pool as shown in Fig.19 As shown in c6, the coefficient of the quadratic term of the curve equation is -0.01045, which is the detection value of the current shape characteristic parameter of the molten pool tail a i is -0.01045. In addition, the position of the tail edge of the fitted molten pool in the original welding area image is as follows: Fig.19 As shown in c7, it can be seen that the shape and position of the edge of the molten pool tail after fitting also reflect the state of the molten pool tail in the original welding area image.

[0097] Under the test conditions of Examples 1 to 3, the double-window interception method and the single-window interception method of the present invention were respectively used to perform real-time sensing detection on the penetration state of the 20 mm long weld during the shaking arc backing welding, and the obtained sampling value a of the characteristic parameter of the tail shape of the molten pool was si Respectively Fig. 20 , 21 , 22. From Fig. 20 , 21 As can be seen from Figure 22, when the penetration state of the base weld is different, the sampling values ​​of the characteristic parameters of the tail shape of the molten pool detected by the two methods are a si The ranges are relatively close. Fig. 20 As shown in the figure, when the bottom weld is fully penetrated, the sampling value a of the characteristic parameter of the tail shape of the molten pool obtained by the double window interception method is si The range is -0.00888~-0.00722, and the sampling value a of the characteristic parameter of the tail shape of the molten pool obtained by the single window interception method si The range is -0.00886 to -0.00717; Fig.21 As shown in the figure, when the base weld is not fully penetrated, the sampling value a of the characteristic parameter of the tail shape of the molten pool obtained by the double window interception method is si The range is -0.00697~-0.00504, and the sampling value a of the characteristic parameter of the tail shape of the molten pool obtained by the single window interception method si The range is -0.0069 to -0.00506; Fig. 22 As shown in the figure, when the base weld is penetrated, the sampling value a of the characteristic parameter of the tail shape of the molten pool obtained by the double window interception method is si The range is -0.01109~-0.009, and the sampling value a of the shape characteristic parameter of the tail of the molten pool obtained by the single window interception method si The range is -0.01125~-0.00912. Correspondingly, when the shaking arc backing weld is not penetrated, the criterion a max Take -0.007, shaking arc backing weld penetration criterion a min When -0.009 is taken, the penetration status of the shaking arc base weld can be accurately determined.

[0098] In addition, for the penetration detection of ordinary arc bottom welds without shaking or swinging arc, the present invention can also quickly and accurately identify the penetration state of the bottom weld by detecting the welding area image during the arc short circuit. Therefore, the bottom weld penetration sensing detection method of the present invention has strong applicability.

Claims

1. An infrared visual sensing detection method for penetrating a shaking arc bottom weld, characterized in that The following steps are involved: ① moving the infrared camera (12) together with the shaking arc welding torch toward the front of the groove of the workpiece to be welded and collecting the welding area image, and sending the welding area image to the computer processing module (15); ② The computer processing module (15) processes the welding area image to obtain the maximum circumscribed rectangle (20) of the molten pool contour and extracts the center of gravity E of the maximum circumscribed rectangle (20). i Coordinate value, height value h i and width w i , using a double-window interception method to intercept the left and right image windows (17-1, 17-2) at the tail of the molten pool, or using a single-window interception method to intercept the image window (17-3) at the tail of the molten pool; ③ Processing the small window images captured by the left and right image windows (17-1, 17-2) of the tail of the molten pool or the small window images (17-3) of the tail of the molten pool to obtain the left and right edge images of the tail of the molten pool or the edge image of the tail of the molten pool; For the left and right edge images corresponding to the left and right image small windows (17-1, 17-2) of the tail of the molten pool, the highest points of the left and right edges and the maximum points of the edge curvature correspondingly searched to the left and right from the highest points are extracted respectively, and the edges between the highest points and the maximum points of the curvature in the corresponding left and right edge images are intercepted as the left and right characteristic edges (18-1, 18-2); For the molten pool tail edge image corresponding to the molten pool tail image small window (17-3), in the portion above half of the height of the edge image, the first extreme point of the ordinate of the molten pool tail edge is searched from the left side of the image to the right and from the right side of the image to the left, and in the molten pool tail edge image, the maximum point of edge curvature is searched from the extreme point searched to the right to the left and from the extreme point searched to the left to the right, respectively. The edge between the extreme point searched to the right and the corresponding maximum point of edge curvature searched to the left, and the edge between the extreme point searched to the left and the corresponding maximum point of edge curvature searched to the right are intercepted as the left and right parts of the characteristic edge of the molten pool tail, and together as the characteristic edge of the molten pool tail (19); ④ The left and right characteristic edges (18-1, 18-2) are simultaneously mapped to the global image coordinate system xOy, or the characteristic edge (19) of the tail of the molten pool is mapped to the global image coordinate system xOy, and a quadratic curve fitting method is used to perform a quadratic curve fitting calculation on the left and right characteristic edges (18-1, 18-2) or the characteristic edge (19) of the tail of the molten pool, and the curve equation of the fitting edge of the tail of the molten pool, the corresponding fitting edge in the global image coordinate system xOy, and the quadratic term coefficient in the quadratic curve equation are obtained, and the quadratic term coefficient is used as the current molten pool tail shape feature parameter detection value a i ; ⑤ By detecting the characteristic parameter value a of the tail shape of the current molten pool i Filter to obtain the sampling value a of the shape characteristic parameter of the current molten pool tail si , determine the sampling value a of the current molten pool tail shape characteristic parameter si , if a si ≤a min The current weld is in the weld-through state. min <a si <a max The current weld is in the penetration state. If a si ≥a max The current weld is in an incomplete penetration state, a min is the criterion for the penetration of the swing arc backing weld, a max It is the criterion for judging whether the shaking arc base weld is not fully penetrated.

2. The infrared visual sensing detection method for the penetration of the shaking arc bottom weld according to claim 1 is characterized in that: In step ②, the double window interception method is as follows: firstly, the end point P of the welding wire is extracted from the inside of the largest circumscribed rectangle (20); i Coordinate value (x Pi ,y Pi ), then according to the center of gravity E of the maximum circumscribed rectangle (20), i Coordinate value (x Ei ,y Ei ), height value h i and width w i The change of respectively adjusts the positioning point B of the left and right image windows (17-1, 17-2) at the tail of the molten pool i , C i The coordinate values ​​are based on the centroid E of the maximum circumscribed rectangle 20 i Coordinate value, height value h i and width w i And the end point P of the welding wire i The change of coordinate values ​​adjusts the height and width of the small image windows (17-1, 17-2) on the left and right sides of the tail of the molten pool.

3. The infrared visual sensing detection method for the penetration of the shaking arc bottom weld according to claim 1 is characterized in that: In step ②, the single window interception method is: according to the center of gravity E of the maximum circumscribed rectangle (20) i Coordinate value (x Ei ,y Ei ), height value h i and width w i Adjust the positioning point F of the small image window (17-3) at the tail of the molten pool by changing i The coordinate values ​​of the maximum circumscribed rectangle (20) are calculated based on the height h i and width w i , adjust the height and width of the small window (17-3) for capturing the image of the tail of the molten pool.

4. The infrared visual sensing detection method for the penetration of the shaking arc bottom weld according to claim 2 is characterized in that: The welding wire end point P i The coordinate extraction method is as follows: on the global binary image of the melt pool, by taking the center of gravity E of the maximum circumscribed rectangle (20) i Search for the pixel with a gray value of 0 and the smallest vertical coordinate value within the 80 pixels × 80 pixels area as the center, and use the coordinates of the pixel as the end point P of the welding wire. i Coordinate (x Pi ,y Pi ).

5. The infrared visual sensing detection method for penetration of a shaking arc bottom weld according to claim 2 is characterized in that: The method for determining the left and right image windows (17-1, 17-2) of the tail of the molten pool is as follows: according to the center of gravity E of the maximum circumscribed rectangle (20) i Coordinate value, height value h i and width w i Determine the positioning point B of the left and right image windows (17-1, 17-2) at the tail of the molten pool i The coordinate value of C i The horizontal coordinate value of the molten pool tail is set to the left and right image windows (17-1, 17-2) at point B. i , C i The vertical coordinate values ​​are equal.

6. The infrared visual sensing detection method for the penetration of the swing arc bottom weld according to claim 4 is characterized in that: When the end point P of the welding wire i The horizontal axis value x Pi Less than or equal to the center of gravity E of the largest circumscribed rectangle (20) i The horizontal axis value x Ei When , the width of the small image window (17-1) on the left side of the tail of the molten pool is w i / 2-|x Pi -x Ei |、Height value is h i / 2+|y Pi -y Ei |, the width of the small window (17-2) of the image capture on the right side of the molten pool tail is w i / 2+|x Pi -x Ei |、Height value is h i / 2-|y Pi -y Ei |; When the end point P of the welding wire i The horizontal axis value x Pi Greater than the center of gravity E of the largest circumscribed rectangle (20) i The horizontal axis value x Ei When , the width of the small window (17-1) for capturing the image on the left side of the tail of the molten pool is w i / 2+|x Pi -x Ei |、Height value is h i / 2-|y Pi -y Ei |, the width of the small window (17-2) of the image capture on the right side of the molten pool tail is w i / 2-|x Pi -x Ei |、Height value is h i / 2+|y Pi -y Ei |.

7. The infrared visual sensing detection method for the penetration of the shaking arc bottom weld according to claim 1 is characterized in that: In step ②, the computer processing module (15) performs global adaptive threshold segmentation on the welding area image using the Otsu method to obtain a global binary image of the molten pool, and detects the maximum circumscribed rectangle (20) of the molten pool contour based on the global binary image of the molten pool.

8. The infrared visual sensing detection method for penetration of a shaking arc bottom weld according to claim 1 is characterized in that: In step ③, the left and right edge images of the molten pool tail (17-1, 17-2) or the small window images captured by the molten pool tail image small window (17-3) are linearly stretched, median filtered, Otsu's adaptive threshold segmentation, and Canny edge extracted to obtain the left and right edge images of the molten pool tail or the molten pool tail edge image.

9. The infrared visual sensing detection method for penetration of a shaking arc bottom weld according to claim 1, characterized in that: In step ⑤, the shape characteristic parameter value a of the current molten pool tail is detected i When filtering, the most recent n≥1 times of the molten pool tail shape characteristic parameter detection value a i The median or average value of is taken as the current sampling value a of the shape characteristic parameter of the tail of the molten pool si .

10. A device for implementing the infrared visual sensing detection method according to any one of claims 1 to 9, characterized in that: The invention comprises a shaking arc welding torch, a voltage sensor (9), a shaking arc welding controller (10), a camera trigger module (11), an infrared camera (12) and a computer processing module (15); the shaking arc welding torch comprises a motor drive and a power feeding mechanism (6), the motor drive and the power feeding mechanism (6) detect the instantaneous shaking position of the welding arc in the groove of the workpiece to be welded relative to the center of the shaking arc welding torch, the shaking arc welding controller (10) controls the motor on the motor drive and the power feeding mechanism (6) to rotate, drives the welding arc at the end of the welding wire to perform periodic reciprocating circular arc trajectory motion around the center of the shaking arc welding torch, and stops at the left and right sides of the groove of the workpiece to be welded respectively, and receives the instantaneous shaking position signal of the arc, and detects the shaking position of the welding arc. The signal is input into the camera trigger module (11); the infrared camera (12) focuses on the welding pool below the welding arc from the front of the groove of the workpiece to be welded at a bird's-eye view shooting angle; in each arc shaking cycle during the shaking arc root welding process, when the welding arc stops at the left or right side of the groove of the workpiece to be welded and shakes to the right or left side within an angle range, the shaking arc welding controller (10) outputs an arc shaking position signal to the camera trigger module (11), and the voltage sensor (9) inputs the detected arc short-circuit voltage signal into the camera trigger module (11); when the camera trigger module (11) detects the first arc short-circuit voltage signal that is less than or equal to the reference voltage, the camera trigger module (11) triggers the infrared camera (12) to collect the welding area image at that moment.

Images