CN113828892B - HDR image-based molten pool center identification system and weld joint tracking method - Google Patents

HDR image-based molten pool center identification system and weld joint tracking method Download PDF

Info

Publication number
CN113828892B
CN113828892B CN202111273178.XA CN202111273178A CN113828892B CN 113828892 B CN113828892 B CN 113828892B CN 202111273178 A CN202111273178 A CN 202111273178A CN 113828892 B CN113828892 B CN 113828892B
Authority
CN
China
Prior art keywords
molten pool
center
wavelet
edge
weld
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202111273178.XA
Other languages
Chinese (zh)
Other versions
CN113828892A (en
Inventor
刘汉钧
石永华
刘志忠
王子顺
叶雄越
钟少涛
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
GUANGDONG FUWEIDE WELDING CO Ltd
Original Assignee
GUANGDONG FUWEIDE WELDING CO Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by GUANGDONG FUWEIDE WELDING CO Ltd filed Critical GUANGDONG FUWEIDE WELDING CO Ltd
Priority to CN202111273178.XA priority Critical patent/CN113828892B/en
Publication of CN113828892A publication Critical patent/CN113828892A/en
Application granted granted Critical
Publication of CN113828892B publication Critical patent/CN113828892B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B23—MACHINE TOOLS; METAL-WORKING NOT OTHERWISE PROVIDED FOR
    • B23K—SOLDERING OR UNSOLDERING; WELDING; CLADDING OR PLATING BY SOLDERING OR WELDING; CUTTING BY APPLYING HEAT LOCALLY, e.g. FLAME CUTTING; WORKING BY LASER BEAM
    • B23K9/00—Arc welding or cutting
    • B23K9/16—Arc welding or cutting making use of shielding gas
    • B23K9/167—Arc welding or cutting making use of shielding gas and of a non-consumable electrode
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B23—MACHINE TOOLS; METAL-WORKING NOT OTHERWISE PROVIDED FOR
    • B23K—SOLDERING OR UNSOLDERING; WELDING; CLADDING OR PLATING BY SOLDERING OR WELDING; CUTTING BY APPLYING HEAT LOCALLY, e.g. FLAME CUTTING; WORKING BY LASER BEAM
    • B23K9/00—Arc welding or cutting
    • B23K9/02—Seam welding; Backing means; Inserts
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B23—MACHINE TOOLS; METAL-WORKING NOT OTHERWISE PROVIDED FOR
    • B23K—SOLDERING OR UNSOLDERING; WELDING; CLADDING OR PLATING BY SOLDERING OR WELDING; CUTTING BY APPLYING HEAT LOCALLY, e.g. FLAME CUTTING; WORKING BY LASER BEAM
    • B23K9/00—Arc welding or cutting
    • B23K9/06—Arrangements or circuits for starting the arc, e.g. by generating ignition voltage, or for stabilising the arc
    • B23K9/067—Starting the arc
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B23—MACHINE TOOLS; METAL-WORKING NOT OTHERWISE PROVIDED FOR
    • B23K—SOLDERING OR UNSOLDERING; WELDING; CLADDING OR PLATING BY SOLDERING OR WELDING; CUTTING BY APPLYING HEAT LOCALLY, e.g. FLAME CUTTING; WORKING BY LASER BEAM
    • B23K9/00—Arc welding or cutting
    • B23K9/12—Automatic feeding or moving of electrodes or work for spot or seam welding or cutting
    • B23K9/127—Means for tracking lines during arc welding or cutting
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B23—MACHINE TOOLS; METAL-WORKING NOT OTHERWISE PROVIDED FOR
    • B23K—SOLDERING OR UNSOLDERING; WELDING; CLADDING OR PLATING BY SOLDERING OR WELDING; CUTTING BY APPLYING HEAT LOCALLY, e.g. FLAME CUTTING; WORKING BY LASER BEAM
    • B23K9/00—Arc welding or cutting
    • B23K9/24—Features related to electrodes
    • B23K9/28—Supporting devices for electrodes
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B23—MACHINE TOOLS; METAL-WORKING NOT OTHERWISE PROVIDED FOR
    • B23K—SOLDERING OR UNSOLDERING; WELDING; CLADDING OR PLATING BY SOLDERING OR WELDING; CUTTING BY APPLYING HEAT LOCALLY, e.g. FLAME CUTTING; WORKING BY LASER BEAM
    • B23K9/00—Arc welding or cutting
    • B23K9/32—Accessories
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B23—MACHINE TOOLS; METAL-WORKING NOT OTHERWISE PROVIDED FOR
    • B23K—SOLDERING OR UNSOLDERING; WELDING; CLADDING OR PLATING BY SOLDERING OR WELDING; CUTTING BY APPLYING HEAT LOCALLY, e.g. FLAME CUTTING; WORKING BY LASER BEAM
    • B23K9/00—Arc welding or cutting
    • B23K9/32—Accessories
    • B23K9/325—Devices for supplying or evacuating shielding gas
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B25—HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
    • B25J—MANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
    • B25J11/00—Manipulators not otherwise provided for
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B25—HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
    • B25J—MANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
    • B25J9/00—Programme-controlled manipulators
    • B25J9/0081—Programme-controlled manipulators with master teach-in means
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B25—HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
    • B25J—MANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
    • B25J9/00—Programme-controlled manipulators
    • B25J9/16—Programme controls
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B25—HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
    • B25J—MANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
    • B25J9/00—Programme-controlled manipulators
    • B25J9/16—Programme controls
    • B25J9/1694—Programme controls characterised by use of sensors other than normal servo-feedback from position, speed or acceleration sensors, perception control, multi-sensor controlled systems, sensor fusion
    • B25J9/1697—Vision controlled systems

Landscapes

  • Engineering & Computer Science (AREA)
  • Mechanical Engineering (AREA)
  • Physics & Mathematics (AREA)
  • Plasma & Fusion (AREA)
  • Robotics (AREA)
  • Manipulator (AREA)

Abstract

The invention discloses a molten pool center identification system and a weld joint tracking method based on HDR images, and solves the problem that in the prior art, the identification precision of a molten pool center and a weld joint is low. The invention uses baud features and gradient information as identification features, and uses a K-means algorithm to screen the features, thereby improving the algorithm efficiency. The method realizes the welding line tracking by acquiring the deviation between the center of the molten pool and the welding line and correcting the position of the K-TIG welding gun in real time according to the deviation, and can be used for the automatic tracking operation of the K-TIG welding. The method has strong robustness, high precision of identifying the center of the molten pool, and improved calculation precision of welding deviation, higher precision of welding seam tracking, and improved automation degree of welding.

Description

HDR image-based molten pool center identification system and weld joint tracking method
Technical Field
The invention relates to the technical field of high dynamic vision, in particular to a molten pool center identification system based on HDR images and a weld joint tracking method.
Background
The keyhole deep-melting TIG welding (K-TIG welding) is a novel efficient welding method for realizing large melting depth by using a keyhole effect in the welding process, can realize one-pass penetration without beveling and single-side welding double-side forming, has good double-side welding line forming and high welding quality, and has wide application prospect in the welding scene of medium-thickness plates. The K-TIG welding is stable and has no splashing, but the heat input of the K-TIG welding is large, so that the plate is easy to generate thermal deformation, and the welding track deviation is caused. In the prior art, the welding track deviation can be calculated through teaching and reproducing, but the welding track deviation caused by thermal deformation cannot be corrected in real time, manual handheld welding is not supported due to large welding current, and automation and intellectualization of welding are realized by adopting a robot welding technology.
At present, sensors such as a visual sensor, an arc sensor, an ultrasonic sensor, an infrared sensor and a mechanical sensor appear in the field of identifying and tracking of welding seams, the advantages and the disadvantages of various sensors are different in adaptation occasions, and the specific application scene is fully considered before the sensors are used. The visual sensor has the advantages of high precision, non-contact type, simple working principle, capability of acquiring a large amount of information from an image, no influence of material types and the like, and can realize the identification and tracking of various welding joints.
The visual sensing technology is divided into active vision and passive vision, in the process of seam tracking, the laser seam tracking technology in the active vision is commercialized, and the laser seam tracking technology has the advantages of high robustness, high precision and the like. For K-TIG welding and plasma arc welding with large heat input, the laser welding seam tracking technology cannot accurately make up for welding deviation caused by thermal deformation. The passive vision refers to a vision technology which does not add an auxiliary light source and takes arc light and natural light as image light sources, and has the advantages of low cost, easiness in installation, full closed-loop control and the like, but the characteristic recognizability of the passive vision is low, and the traditional image processing method is difficult to accurately recognize the center of a molten pool and a welding seam.
The invention patent CN 110238477A discloses a mechanically rotating weld joint tracking system for laser brazing and a working method thereof, welding deviation is obtained by adopting welding wire sensing, and the deviation is corrected by cylinder control. However, this patent does not explain a specific recognition method, and since the K-TIG welding uses a tungsten needle for welding and does not have its sensing condition, it is difficult to apply the patent to the field of K-TIG welding.
The invention patent CN 111299760A discloses a robot weld joint tracking and weld pool monitoring sensor based on active and passive vision, wherein laser weld joint tracking and weld pool monitoring are combined in one sensor, laser is irradiated in a V-shaped weld joint, and the actual position of the weld joint is calculated by identifying the turning point of the laser. The laser welding seam tracking has high robustness and high precision, but the cost is high. The K-TIG welding is normally not beveled and is commonly used for welding narrow-gap workpieces, so that the laser spot characteristics are not as obvious as those of a V-shaped welding seam, and the problem of difficult identification exists.
In order to solve the problem of difficulty in identifying a weld joint and a molten pool by passive vision, an intelligent identification method capable of accurately extracting the edge of the weld joint and the edge of the molten pool needs to be designed, and meanwhile, automatic weld joint tracking of robot K-TIG welding is achieved.
Disclosure of Invention
Aiming at the defects in the prior art, the weld pool center identification system and the weld pool tracking method based on the HDR image solve the problem that the weld and the weld pool are difficult to identify in the prior art.
In order to achieve the purpose of the invention, the invention adopts the technical scheme that: a molten pool center identification system based on HDR images comprises an industrial robot, an industrial robot control cabinet, an industrial robot demonstrator, an HDR camera, a fixed arm, an image acquisition card, a deep melting K-TIG welding power supply, a water cooler, a K-TIG welding gun, an argon gas bottle, a PC (personal computer) and an OPC (OLE for process control) communication lower computer;
the tail end of a motion execution mechanism of the industrial robot is clamped with a K-TIG welding gun, the industrial robot is connected with an industrial robot control cabinet, and the industrial robot control cabinet is respectively connected with an industrial robot demonstrator and an OPC communication lower computer;
the HDR camera is arranged on the fixing arm, an included angle between the axis of the HDR camera and the axis of the K-TIG welding gun is 60 degrees, and the axis of the HDR camera and the axis of the K-TIG welding gun are coplanar; the fixed arm is arranged on a motion executing mechanism of the industrial robot, the HDR camera is connected with an image acquisition card, the image acquisition card is connected with a PC (personal computer), and the PC is connected with an OPC (OLE for process control) communication lower computer;
the K-TIG welding gun is respectively connected with the water cooling machine and the deep melting K-TIG welding power supply, and the deep melting K-TIG welding power supply is respectively connected with the argon gas bottle and the OPC communication lower computer.
Further, the industrial robot is a six-axis industrial robot, and the fixed arm is a three-degree-of-freedom mechanical arm.
A method of seam tracking using an HDR image-based puddle center identification system, comprising:
calibrating the HDR camera to obtain a conversion relation matrix of a pixel coordinate system and a world coordinate system;
acquiring an HDR welding region image through an HDR camera, extracting a region of interest with fixed resolution in the HDR welding region image, and performing multi-scale wavelet transformation on the region of interest to obtain a wavelet transformation result, wherein the wavelet transformation result comprises 3n '+1 wavelet features corresponding to each pixel in the region of interest, and n' represents the degree of scale;
performing linear interpolation on the wavelet transformation result to obtain a wavelet characteristic matrix;
acquiring gradient information of the region of interest by using a Sobel operator;
screening wavelet features in the wavelet feature matrix by adopting a K-means clustering algorithm to obtain N wavelet features with the maximum identification degree, and obtaining weld joint identification features through the N wavelet features with the maximum identification degree and gradient information;
acquiring a label image through the region of interest, and training two support vector machines according to the label image and the weld joint identification characteristics to respectively obtain an HDR welding region molten pool edge identification model and a weld joint edge identification model;
identifying the region of interest to be identified through an HDR welding region molten pool edge identification model and a weld joint edge identification model respectively to obtain a molten pool edge and a weld joint edge;
fitting an elliptic equation of the edge of the molten pool and a linear equation of the edge of the fitted weld seam according to the edge of the molten pool and the edge of the weld seam, obtaining the center of the molten pool according to the obtained elliptic equation, obtaining a vertical point from the center of the molten pool to the central line of the weld seam, and obtaining a coordinate of the center of the molten pool and a coordinate of the vertical point;
the central point of the ellipse equation is the center of a molten pool, and the linear equation is the central line of a welding seam;
respectively converting the coordinate of the center of the molten pool and the coordinate of the vertical point into world coordinates according to the conversion relation matrix of the pixel coordinate system and the world coordinate system, respectively obtaining the world coordinate of the center of the molten pool and the world coordinate of the vertical point, and obtaining the deviation between the center of the molten pool and the vertical point according to the world coordinate of the center of the molten pool and the world coordinate of the vertical point;
and controlling the industrial robot to correct the deviation through the PC according to the deviation.
Further, the conversion relation matrix of the pixel coordinate system and the world coordinate system is as follows:
Figure BDA0003329329410000041
wherein (X)w,Yw,Zw) Respectively representing coordinates on an X axis, a Y axis and a Z axis in a world coordinate system, R representing an external rotation matrix, T representing a translation matrix, f representing the focal length of the HDR camera, and dx and dy representing the actual distances of one pixel point in the HDR welding area image on the X axis and the Y axis in a pixel coordinate system; u. of0And v0Respectively representing the offset of the origin of the HDR welding region image and the origin of the camera coordinate on an x axis and a y axis under a pixel coordinate system; (X)p,Yp) Representing the coordinates on the x-axis and the y-axis, respectively, in the pixel coordinate system.
Further, the acquiring gradient information of the region of interest by using the Sobel operator includes:
carrying out convolution operation on the region of interest by adopting convolution factors of Sobel operator to obtain gradient G in the x directionxAnd gradient G in y-directionyThe convolution factor of the Sobel operator is:
Figure BDA0003329329410000051
according to the gradient G of the x directionxAnd gradient G in y-directionyAcquiring gradient information of the region of interest:
Figure BDA0003329329410000052
Figure BDA0003329329410000053
where G denotes the gradient magnitude and θ denotes the gradient direction.
Further, the screening of the wavelet features in the wavelet feature matrix by using the K-means clustering algorithm to obtain N wavelet features with the maximum degree of identification, and obtaining the weld joint identification features through the N wavelet features with the maximum degree of identification and the gradient information includes:
combining gradient information, taking the gradient size, the gradient space and the first wavelet feature corresponding to each pixel in the wavelet feature matrix as feature spaces, and clustering the feature spaces corresponding to each pixel based on a K-means clustering algorithm to obtain a K-means clustering result, wherein the K value of the K-means clustering algorithm is 3, and the K-means clustering result comprises the classification of each pixel;
obtaining the intra-class difference of each class according to the K-means clustering result
Figure BDA0003329329410000054
Comprises the following steps:
Figure BDA0003329329410000055
wherein i represents the category of the K-means clustering result, the value range of i is {1,2,3}, {1,2,3} respectively represents the edge of the molten pool, the welding line and the background, n represents the number of pixel points contained in the category i, and X represents the number of the pixel points contained in the category ijThe wavelet feature value corresponding to the j-th pixel point in the category i is represented, wherein j is 1,2, …, n,
Figure BDA0003329329410000056
representing the mean value of wavelet characteristic values corresponding to all pixel points in the category i;
obtaining the difference between the two categories according to the K-means clustering result
Figure BDA0003329329410000057
Comprises the following steps:
Figure BDA0003329329410000061
wherein i ' represents the category of the K-means clustering result, the value range of i ' is {1,2,3}, i ' represents the category different from i,
Figure BDA0003329329410000062
representing the mean value of wavelet characteristic values corresponding to all pixel points in the category i';
according to the difference in class
Figure BDA0003329329410000063
And difference in
Figure BDA0003329329410000064
The identification CS for obtaining the first wavelet feature is:
Figure BDA0003329329410000065
wherein,
Figure BDA0003329329410000066
an intra-class difference representing a class i';
based on the steps, traversing 3n '+1 wavelet features corresponding to each pixel to obtain 3n' +1 identification degrees;
and acquiring N wavelet features with the maximum identification degree corresponding to each pixel point according to the identification degree of the wavelet features, and forming the weld joint identification features by the N wavelet features with the maximum identification degree and the gradient information.
Further, the support vector machine employs a gaussian kernel function.
Further, the fitting an elliptic equation of the edge of the molten pool and a linear equation of the edge of the fitted weld according to the edge of the molten pool and the edge of the weld, obtaining the center of the molten pool according to the obtained elliptic equation, obtaining a vertical point from the center of the molten pool to the central line of the weld, and obtaining a coordinate of the center of the molten pool and a coordinate of the vertical point comprises:
the initial ellipse equation is constructed as follows:
ax2+bxy+cy2+dx+ey+1＝0
wherein (x, y) represents coordinates of a pixel coordinate system, and a, b, c, d, and e represent a first coefficient, a second coefficient, a third coefficient, a fourth coefficient, and a fifth coefficient, respectively;
randomly selecting M points in the edge of the molten pool, solving the coefficient of an initial elliptic equation according to the coordinates of the M points to obtain an iterative elliptic equation, wherein M is more than or equal to 5, any three points in the M points are not collinear, and the iterative elliptic equation is as follows:
amx2+bmxy+cmy2+dmx+emy+1＝0
wherein m represents the number of iterations, am、bm、cm、dmAnd emRespectively representing the solved values of a, b, c, d and e of the mth iteration;
traversing each point in the edge of the molten pool, and acquiring the deviation sigma between each point in the edge of the molten pool and the iterative elliptic equation as follows:
σ＝|fn-1|
Figure BDA0003329329410000071
wherein f isnDenotes the intermediate parameter, (x)n,yn) The coordinate of the nth point in the edge of the molten pool is represented, wherein n is 1,2, …, L and L represents the total number of the middle points of the edge of the molten pool;
the iteration times m are set to be 1, the steps are repeated, the counting value of the iteration times m in each repeated process is added with one until the iteration times m reach a set threshold value, an iteration elliptic equation with the largest number of inner points is selected as an elliptic equation of the edge of the molten pool, and the inner points refer to points with the deviation sigma smaller than 5;
according to an elliptic equation, the center of the molten pool is obtained as follows:
Figure BDA0003329329410000072
Figure BDA0003329329410000073
wherein (x)c,yc) Coordinates representing the center of the molten pool;
the linear equation of the welding seam edge is fitted by adopting a least square method as follows:
y＝gx+h
Figure BDA0003329329410000074
Figure BDA0003329329410000075
wherein g represents a sixth coefficient, h represents a seventh coefficient, (x)q,yq) Coordinates of the Q-th point in the edge of the weld are shown, Q is 1,2, …, Q represents the total number of the middle points of the edge of the weld,
Figure BDA0003329329410000076
representing the mean value of coordinates of all points in the edge of the weld;
obtaining the coordinate (x) of the vertical point from the center of the molten pool to the center line of the welding seamv,yv) Comprises the following steps:
Figure BDA0003329329410000081
further, the converting the coordinate of the center of the molten pool and the coordinate of the vertical point into a world coordinate according to the conversion relation matrix of the pixel coordinate system and the world coordinate system, respectively obtaining the world coordinate of the center of the molten pool and the world coordinate of the vertical point, and obtaining the deviation between the center of the molten pool and the vertical point according to the world coordinate of the center of the molten pool and the world coordinate of the vertical point, includes:
according to the conversion relation moment of the pixel coordinate system and the world coordinate systemThe array converts the coordinate of the center of the molten pool and the coordinate of the vertical point into world coordinates, and respectively obtains the world coordinates of the center of the molten pool and the world coordinates of the vertical point as
Figure BDA0003329329410000082
And
Figure BDA0003329329410000083
acquiring the deviation between the center of the molten pool and the vertical point according to the world coordinates of the center of the molten pool and the world coordinates of the vertical point:
Figure BDA0003329329410000084
Figure BDA0003329329410000085
wherein σXRepresenting the deviation between the centre of the bath and the vertical point in the X-axis, σYIndicating the deviation between the center of the bath and the vertical point on the Y-axis.
Further, according to the deviation, the deviation correction is carried out by controlling the industrial robot through the PC machine, and the method comprises the following steps: and filtering the deviation through a Kalman filter to obtain a final welding deviation, and controlling the industrial robot to correct the deviation through the PC according to the final welding deviation.
The invention has the beneficial effects that:
(1) the invention provides a molten pool center identification system and a weld joint tracking method based on HDR images, which solve the problem of low identification precision of a molten pool center and a weld joint in the prior art.
(2) The method has strong robustness, high precision of identifying the center of the molten pool, and improved calculation precision of welding deviation, higher precision of welding seam tracking and improved automation degree of welding.
(3) The invention has high real-time performance and simple and convenient application, filters wavelet characteristics by a k-means clustering algorithm, can accurately adjust the wavelet transformation depth according to the filtering result, reduces the computing resource and time consumed by wavelet transformation, and can quickly and effectively filter the characteristics because the k-means clustering algorithm belongs to unsupervised learning and does not relate to the problem of identification model training.
Drawings
Fig. 1 is a block diagram of a HDR image-based molten pool center identification system according to an embodiment of the present invention.
FIG. 2 is a flowchart of a weld tracking method according to an embodiment of the present invention.
Wherein: the system comprises an industrial robot 1, an industrial robot 2, an industrial robot control cabinet 3, an industrial robot demonstrator 4-HDR camera, a fixed arm 5, an image acquisition card 6, a deep melting K-TIG welding power supply 7, a water cooling machine 8, a TIG welding gun 9, an argon gas bottle 10, a PC 11 and an OPC 12 communication lower computer.
Detailed Description
The following description of the embodiments of the present invention is provided to facilitate the understanding of the present invention by those skilled in the art, but it should be understood that the present invention is not limited to the scope of the embodiments, and it will be apparent to those skilled in the art that various changes may be made without departing from the spirit and scope of the invention as defined and defined in the appended claims, and all matters produced by the invention using the inventive concept are protected.
Embodiments of the present invention will be described in detail below with reference to the accompanying drawings.
As shown in figure 1, the molten pool center identification system based on HDR (high dynamic illumination rendering) images comprises an industrial robot 1, an industrial robot control cabinet 2, an industrial robot demonstrator 3, an HDR camera 4, a fixed arm 5, an image acquisition card 6, a deep melting K-TIG welding power supply 7, a water cooler 8, a K-TIG welding gun 9, an argon gas bottle 10, a PC (personal computer) 11 and an OPC (optical proximity correction) communication lower computer 12.
An industrial robot 1, an industrial robot Control cabinet 2 and an industrial robot demonstrator 3 form a welding gun motion Control module, an HDR camera 4, a fixed arm 5 and an image acquisition card 6 form an HDR image sensing module, a deep melting K-TIG (Keyhole Tungsten Inert Gas) welding power supply 7, a water cooler 8, a K-TIG welding gun 9 and an argon bottle 10 form a welding energy input module, and a PC (computer) machine 11 and an OPC (OLE for Process Control, object linking and embedded Process Control) communication lower computer 12 form an identification Control module. The welding gun motion control module can be used for controlling the motion track of the welding gun, the HDR image sensing module is used for collecting HDR images, the welding energy input module is used for welding along the motion track of the welding gun, the identification control module is used for controlling the whole system to work, the welding deviation in the HDR images is identified, a deviation correction instruction is sent to the welding gun motion control module, and deviation correction is carried out.
The tail end of a motion executing mechanism of the industrial robot 1 is clamped with a K-TIG welding gun 9, the industrial robot 1 is connected with an industrial robot control cabinet 2, and the industrial robot control cabinet 2 is respectively connected with an industrial robot demonstrator 3 and an OPC communication lower computer 12.
The HDR camera 4 is arranged on the fixing arm 5, an included angle between the axis of the HDR camera 4 and the axis of the K-TIG welding gun 9 is 60 degrees, and the axis of the HDR camera 4 and the axis of the K-TIG welding gun 9 are coplanar; the fixed arm 5 is arranged on a motion executing mechanism of the industrial robot 1, the HDR camera 4 is connected with the image acquisition card 6, the image acquisition card 6 is connected with the PC 11, and the PC 11 is connected with the OPC communication lower computer 12.
The K-TIG welding gun 9 is respectively connected with the water cooling machine 8 and the deep melting K-TIG welding power supply 7, and the deep melting K-TIG welding power supply 7 is respectively connected with the argon gas bottle 10 and the OPC communication lower computer 12.
In one possible embodiment, the industrial robot 1 is a six-axis industrial robot.
In a possible embodiment, said fixed arm 5 is a three-degree-of-freedom mechanical arm.
The control method of the molten pool center identification system comprises the following steps:
step 1: after the system is powered on, the argon bottle 10 is opened to ensure the supply of argon.
Step 2: the industrial robot 1 is controlled to reach the welding start position by the industrial robot demonstrator 3 and the industrial robot control cabinet 2.
And step 3: the image acquisition card 6 and the OPC communication lower computer 12 are connected through the PC 11, so that the PC 11 is in an image acquisition state and a control state, and proper sampling frequency and robot movement speed are set.
And 4, step 4: and setting a proper welding current for the deep melting K-TIG welding power supply 7.
And 5: the PC 11 controls the deep-melting K-TIG welding power supply 7 to start arc through the OPC communication lower computer 12, and the water cooler 8 ensures that the K-TIG welding gun 9 is not overheated in the welding process.
Step 6: the method comprises the steps of controlling the industrial robot 1 to move in a welding direction at a certain speed through the PC 11, collecting HDR images through the HDR camera 4, transmitting the HDR images to the PC 11 through the image collecting card 6, extracting welding deviation through the PC 11, and controlling the industrial robot 1 to correct the deviation according to the welding deviation.
And 7: after welding is finished, controlling the deep-melting K-TIG welding power supply 7 to extinguish arc through the PC 11, controlling the industrial robot 1 to stop moving after arc extinguishing, and disconnecting the PC 11 from the image acquisition card 6 and the OPC communication lower computer 12 after complete stop of movement;
and 8: the system is powered off.
As shown in fig. 2, a weld tracking method using an HDR image-based weld puddle center identification system includes:
A. the HDR camera 4 is calibrated, and the PC 11 acquires a conversion relation matrix between the pixel coordinate system and the world coordinate system.
In the present embodiment, the HDR camera 4 is calibrated by the zhangnyou calibration method.
B. An HDR welding region image is collected through an HDR camera 4, a region of interest with fixed resolution in the HDR welding region image is extracted, multi-scale wavelet transformation is carried out on the region of interest, a wavelet transformation result is obtained, the wavelet transformation result comprises 3n '+1 wavelet features corresponding to each pixel in the region of interest, and n' represents the degree of scale.
The wavelet transform of each scale can obtain four wavelet features, namely a low-frequency component, X, Y and a diagonal high-frequency feature, wherein the low-frequency component is further subjected to wavelet transform of the next scale, that is, the number of the wavelet features obtained finally is 3N '+1, where N is the depth of the wavelet transform, that is, how many scales are divided for wavelet transform, and the wavelet feature at each pixel position is a vector of 3N' + 1.
In this embodiment, the fixed resolution is set to 800 × 800, a region of interest with the fixed resolution of 800 × 800 in the HDR welding region image is extracted, and a multi-scale wavelet transform with a Harr wavelet (haar wavelet) as a wavelet basis and a depth of 4 is performed on the region of interest to obtain a wavelet transform result.
C. And performing linear interpolation on the wavelet transform result to obtain a wavelet feature matrix, wherein the dimension of the wavelet feature matrix is n 'xw' xh ', wherein n' is the wavelet feature number, w 'is the width of the region of interest, h' is the height of the region of interest, and the unit of the wavelet feature matrix are pixels.
And performing multi-scale wavelet transformation with the depth of 4 on the region of interest by taking the Harr wavelet as a wavelet base to obtain 12 feature maps, wherein the resolution of the feature images is reduced by a factor of 1/2 with the increase of the scale, and linear interpolation is used for restoring the resolution of the feature maps to be consistent with the region of interest in order to keep the features aligned.
In this embodiment, linear interpolation is performed on the wavelet transform result to restore the resolution to 800 × 800, and a 12 × 800 × 800 wavelet feature matrix is obtained by arranging wavelet features from small to large according to the depth of the wavelet transform and by reverse-time sorting.
D. And acquiring gradient information of the region of interest by using a Sobel operator (Sobel operator).
And solving gradient information of the region of interest by using a Sobel operator, wherein the gradient information comprises a gradient size G and a gradient angle theta, and obtaining a gradient characteristic matrix of 2 x 800.
E. And screening wavelet features in the wavelet feature matrix by adopting a K-means (K mean value) clustering algorithm to obtain N wavelet features with the maximum identification degree, and obtaining the weld joint identification features through the N wavelet features with the maximum identification degree and the gradient information.
In the present embodiment, N is set to 8.
F. And acquiring a label image through the region of interest, and training the two support vector machines according to the label image and the weld joint identification characteristics to respectively obtain an HDR welding region molten pool edge identification model and a weld joint edge identification model.
The label image corresponds to the weld joint identification characteristics, and the weld pool edge and the weld joint edge in the region of interest can be marked in a manual marking mode, so that the label image is obtained. During training, the support vector machine is trained by taking the weld joint identification characteristics as training data and taking the label image as a label of the training data.
G. And identifying the region to be identified through the HDR welding region molten pool edge identification model and the weld joint edge identification model respectively to obtain the molten pool edge and the weld joint edge.
And respectively inputting the weld joint identification characteristics corresponding to the image to be identified into a weld pool edge identification model and a weld joint edge identification model in the HDR welding area to obtain the weld pool edge and the weld joint edge.
H. According to the edge of the molten pool and the edge of the welding seam, fitting an elliptic equation of the edge of the molten pool by adopting a random sampling consensus algorithm, fitting a linear equation of the edge of the welding seam by adopting a least square method, obtaining the center of the molten pool according to the obtained elliptic equation, obtaining a vertical point from the center of the molten pool to the central line of the welding seam, and obtaining the coordinate of the center of the molten pool and the coordinate of the vertical point.
The central point of the ellipse equation is the center of a molten pool, and the linear equation is the central line of a welding seam;
I. and respectively converting the coordinate of the center of the molten pool and the coordinate of the vertical point into world coordinates according to the conversion relation matrix of the pixel coordinate system and the world coordinate system, respectively obtaining the world coordinate of the center of the molten pool and the world coordinate of the vertical point, and acquiring the deviation between the center of the molten pool and the vertical point according to the world coordinate of the center of the molten pool and the world coordinate of the vertical point.
J. According to the deviation, the industrial robot 1 is controlled by the PC machine 11 to correct the deviation.
In a possible embodiment, the control of the industrial robot 1 by the PC machine 11 for deviation correction based on the deviation comprises: and filtering the deviation through a Kalman filter to obtain a final welding deviation, and controlling the industrial robot 1 to correct the deviation through the PC 11 according to the final welding deviation.
In a possible implementation manner, the matrix of the conversion relationship between the pixel coordinate system and the world coordinate system in step a is:
Figure BDA0003329329410000141
wherein (X)w,Yw,Zw) Respectively representing coordinates on an X axis, a Y axis and a Z axis in a world coordinate system, R representing an external rotation matrix, T representing a translation matrix, f representing the focal length of the HDR camera 4, and dx and dy representing the actual distance of one pixel point in the HDR welding area image on the X axis and the Y axis in a pixel coordinate system; u. of0And v0Respectively representing the offset of the origin of the HDR welding region image and the origin of the camera coordinate on an x axis and a y axis under a pixel coordinate system; (X)p,Yp) Representing the coordinates on the x-axis and the y-axis, respectively, in the pixel coordinate system.
In a possible implementation manner, the acquiring, in step D, gradient information of the region of interest using a Sobel operator includes:
carrying out convolution operation on the region of interest by adopting convolution factors of Sobel operator to obtain gradient G in the x directionxAnd gradient G in y-directionyThe convolution factor of the Sobel operator is:
Figure BDA0003329329410000142
according to the gradient G of the x directionxAnd gradient G in y-directionyAcquiring gradient information of the region of interest:
Figure BDA0003329329410000143
Figure BDA0003329329410000144
where G denotes the gradient magnitude and θ denotes the gradient direction.
In a possible implementation, the step E includes:
combining gradient information, taking the gradient size, the gradient space and the first wavelet feature corresponding to each pixel in the wavelet feature matrix as feature spaces, and clustering the feature spaces corresponding to each pixel based on a K-means clustering algorithm to obtain a K-means clustering result, wherein the K value of the K-means clustering algorithm is 3, and the K-means clustering result comprises the classification of each pixel;
obtaining the intra-class difference of each class according to the K-means clustering result
Figure BDA0003329329410000151
Comprises the following steps:
Figure BDA0003329329410000152
wherein i represents the category of the K-means clustering result, the value range of i is {1,2,3}, {1,2,3} respectively represents the edge of the molten pool, the welding line and the background, n represents the number of pixel points contained in the category i, and X represents the number of the pixel points contained in the category ijThe wavelet feature value corresponding to the j-th pixel point in the category i is represented, wherein j is 1,2, …, n,
Figure BDA0003329329410000153
representing the mean value of wavelet characteristic values corresponding to all pixel points in the category i;
obtaining the difference between the two categories according to the K-means clustering result
Figure BDA0003329329410000154
Comprises the following steps:
Figure BDA0003329329410000155
wherein i' represents the category of the K-means clustering result, iThe value range is {1,2,3}, i' represents a different category from i,
Figure BDA0003329329410000156
representing the mean value of wavelet characteristic values corresponding to all pixel points in the category i';
according to the difference in class
Figure BDA0003329329410000157
And difference in
Figure BDA0003329329410000158
The identification CS for obtaining the first wavelet feature is:
Figure BDA0003329329410000159
wherein,
Figure BDA00033293294100001510
an intra-class difference representing a class i';
based on the steps, traversing 3n '+1 wavelet features corresponding to each pixel to obtain 3n' +1 identification degrees;
and acquiring N wavelet features with the maximum identification degree corresponding to each pixel point according to the identification degree of the wavelet features, and forming the weld joint identification features by the N wavelet features with the maximum identification degree and the gradient information.
It should be noted that the identification obtained each time is the identification of the wavelet features corresponding to all the pixel points, for example, the identification obtained for the first time is the identification of the first wavelet feature corresponding to each pixel point, and the identification obtained for the second time is the identification of the second wavelet feature corresponding to each pixel point. Selecting N wavelet features with the largest identification degree, that is, selecting the wavelet features of N pixels with the same serial number, for example, assuming that N is 1, when the identification degree calculated by the first wavelet feature corresponding to all the pixel points is the largest, taking the first wavelet feature as the N wavelet features with the largest identification degree corresponding to each pixel point.
Similar internal differenceDifferent from each other
Figure BDA0003329329410000161
Representing the difference value of the wavelet characteristics in a single category after single clustering; difference in
Figure BDA0003329329410000162
And representing the difference value of the wavelet characteristics between the two categories after single clustering.
After the identification degrees of the 12 wavelet features are respectively calculated, the wavelet features are sorted from large to small, and the 8 wavelet features with the largest identification degrees and the gradient size and the gradient direction in the gradient information are selected to form the weld joint identification feature.
In a possible implementation, the support vector machine in step F employs a gaussian kernel function.
In this embodiment, the gaussian kernel function is specifically:
Figure BDA0003329329410000163
wherein σ ' represents the standard deviation of the gaussian kernel function, which belongs to a hyper-parameter, and since data is normalized before being input into the SVM, the standard deviation of the kernel function takes 1, x "and z ' to represent features of two different dimensions in a feature space, for example, if x" represents wavelet feature 1, z ' can be wavelet feature 2/3/4/…/gradient feature, etc.
In one possible embodiment, the step H comprises:
the initial ellipse equation is constructed as follows:
ax2+bxy+cy2+dx+ey+1＝0
where (x, y) denotes coordinates of a pixel coordinate system, and a, b, c, d, and e denote a first coefficient, a second coefficient, a third coefficient, a fourth coefficient, and a fifth coefficient, respectively.
Randomly selecting M points in the edge of the molten pool, solving the coefficient of an initial elliptic equation according to the coordinates of the M points to obtain an iterative elliptic equation, wherein M is more than or equal to 5, any three points in the M points are not collinear, and the iterative elliptic equation is as follows:
amx2+bmxy+cmy2+dmx+emy+1＝0
wherein m represents the number of iterations, am、bm、cm、dmAnd emThe solution values of a, b, c, d and e for the mth iteration are shown, respectively.
Traversing each point in the edge of the molten pool, and acquiring the deviation sigma between each point in the edge of the molten pool and the iterative elliptic equation as follows:
σ＝|fn-1|
Figure BDA0003329329410000171
wherein, fnDenotes the intermediate parameter, (x)n,yn) The coordinate of the nth point in the edge of the molten pool is shown, wherein n is 1,2, …, L and L is the total number of the middle points of the edge of the molten pool.
And (3) repeating the steps with the iteration number m equal to 1, adding one to the count value of the iteration number m in each repeated process until the iteration number m reaches a set threshold, and selecting an iteration elliptic equation containing the largest number of inner points as an elliptic equation of the edge of the molten pool, wherein the inner points refer to points with the deviation sigma smaller than 5.
According to an elliptic equation, the center of the molten pool is obtained as follows:
Figure BDA0003329329410000172
Figure BDA0003329329410000173
wherein (x)c,yc) Coordinates representing the center of the molten pool.
The linear equation of the welding seam edge is fitted by adopting a least square method as follows:
y＝gx+h
Figure BDA0003329329410000174
Figure BDA0003329329410000175
wherein g represents a sixth coefficient, h represents a seventh coefficient, (x)q,yq) Coordinates of the Q-th point in the edge of the weld are shown, Q is 1,2, …, Q represents the total number of the middle points of the edge of the weld,
Figure BDA0003329329410000181
represents the mean of the coordinates of all points in the weld edge.
Obtaining the coordinate (x) of the vertical point from the center of the molten pool to the center line of the welding seamv,yv) Comprises the following steps:
Figure BDA0003329329410000182
in one possible embodiment, the step I includes:
converting the coordinate of the center of the molten pool and the coordinate of the vertical point into world coordinates according to the conversion relation matrix of the pixel coordinate system and the world coordinate system, and respectively obtaining the world coordinates of the center of the molten pool and the world coordinates of the vertical point as
Figure BDA0003329329410000183
And
Figure BDA0003329329410000184
acquiring the deviation between the center of the molten pool and the vertical point according to the world coordinates of the center of the molten pool and the world coordinates of the vertical point:
Figure BDA0003329329410000185
Figure BDA0003329329410000186
wherein σXRepresenting the deviation between the centre of the bath and the vertical point in the X-axis, σYIndicating the deviation between the center of the bath and the vertical point on the Y-axis.

Claims (10)

1. A molten pool center identification system based on HDR images is characterized by comprising an industrial robot (1), an industrial robot control cabinet (2), an industrial robot demonstrator (3), an HDR camera (4), a fixed arm (5), an image acquisition card (6), a deep melting K-TIG welding power supply (7), a water cooler (8), a K-TIG welding gun (9), an argon gas bottle (10), a PC (11) and an OPC communication lower computer (12);
a K-TIG welding gun (9) is clamped at the tail end of a motion executing mechanism of the industrial robot (1), the industrial robot (1) is connected with an industrial robot control cabinet (2), and the industrial robot control cabinet (2) is respectively connected with an industrial robot demonstrator (3) and an OPC communication lower computer (12);
the HDR camera (4) is arranged on the fixing arm (5), an included angle between the axis of the HDR camera (4) and the axis of the K-TIG welding gun (9) is 60 degrees, and the axis of the HDR camera (4) and the axis of the K-TIG welding gun (9) are coplanar; the fixing arm (5) is arranged on a motion executing mechanism of the industrial robot (1), the HDR camera (4) is connected with the image acquisition card (6), the image acquisition card (6) is connected with the PC (11), and the PC (11) is connected with the OPC communication lower computer (12);
the K-TIG welding gun (9) is respectively connected with a water cooling machine (8) and a deep melting K-TIG welding power supply (7), and the deep melting K-TIG welding power supply (7) is respectively connected with an argon gas bottle (10) and an OPC communication lower computer (12);
the weld tracking method using the HDR image-based molten pool center identification system comprises the following steps:
calibrating the HDR camera (4) to obtain a conversion relation matrix of a pixel coordinate system and a world coordinate system;
acquisition of an HDR weld region map by an HDR camera (4)Extracting a region of interest with fixed resolution in the HDR welding region image, and performing multi-scale wavelet transformation on the region of interest to obtain a wavelet transformation result, wherein the wavelet transformation result comprises a wavelet corresponding to each pixel in the region of interest
Figure 734793DEST_PATH_IMAGE001
The characteristics of the wavelet are shown in the figure,
Figure 284723DEST_PATH_IMAGE002
indicating the degree of the scale;
performing linear interpolation on the wavelet transformation result to obtain a wavelet characteristic matrix;
acquiring gradient information of the region of interest by using a Sobel operator;
screening the wavelet characteristics in the wavelet characteristic matrix by adopting a K-means clustering algorithm to obtain wavelet characteristicsNFeatures of wavelet with maximum discrimination, and passingNAcquiring the weld joint identification characteristics from the small wave characteristics and the gradient information with the maximum identification degree;
acquiring a label image through the region of interest, and training two support vector machines according to the label image and the weld joint identification characteristics to respectively obtain an HDR welding region molten pool edge identification model and a weld joint edge identification model;
identifying the region of interest to be identified through an HDR welding region molten pool edge identification model and a weld joint edge identification model respectively to obtain a molten pool edge and a weld joint edge;
fitting an elliptic equation of the edge of the molten pool and a linear equation of the edge of the fitted weld seam according to the edge of the molten pool and the edge of the weld seam, obtaining the center of the molten pool according to the obtained elliptic equation, obtaining a vertical point from the center of the molten pool to the central line of the weld seam, and obtaining a coordinate of the center of the molten pool and a coordinate of the vertical point;
the central point of the ellipse equation is the center of a molten pool, and the linear equation is the central line of a welding seam;
respectively converting the coordinate of the center of the molten pool and the coordinate of the vertical point into world coordinates according to the conversion relation matrix of the pixel coordinate system and the world coordinate system, respectively obtaining the world coordinate of the center of the molten pool and the world coordinate of the vertical point, and obtaining the deviation between the center of the molten pool and the vertical point according to the world coordinate of the center of the molten pool and the world coordinate of the vertical point;
and controlling the industrial robot (1) to correct the deviation through the PC (11) according to the deviation.
2. The HDR image-based molten pool center identifying system according to claim 1, wherein the industrial robot (1) is a six-axis industrial robot, and the fixed arm (5) is a three-degree-of-freedom mechanical arm.
3. A method of seam tracking using the HDR image-based puddle center identification system of claim 1, comprising:
calibrating the HDR camera (4) to obtain a conversion relation matrix of a pixel coordinate system and a world coordinate system;
an HDR welding region image is collected through an HDR camera (4), a region of interest with fixed resolution in the HDR welding region image is extracted, multi-scale wavelet transformation is carried out on the region of interest to obtain a wavelet transformation result, and the wavelet transformation result comprises a wavelet corresponding to each pixel in the region of interest
Figure 353042DEST_PATH_IMAGE003
The characteristics of the wavelet are shown in the figure,
Figure 912199DEST_PATH_IMAGE004
indicating the degree of the scale;
performing linear interpolation on the wavelet transformation result to obtain a wavelet characteristic matrix;
acquiring gradient information of the region of interest by using a Sobel operator;
screening the wavelet characteristics in the wavelet characteristic matrix by adopting a K-means clustering algorithm to obtain wavelet characteristicsNFeatures of wavelet with maximum discrimination, and passingNAcquiring the weld joint identification characteristics from the small wave characteristics and the gradient information with the maximum identification degree;
acquiring a label image through the region of interest, and training two support vector machines according to the label image and the weld joint identification characteristics to respectively obtain an HDR welding region molten pool edge identification model and a weld joint edge identification model;
identifying the region of interest to be identified through an HDR welding region molten pool edge identification model and a weld joint edge identification model respectively to obtain a molten pool edge and a weld joint edge;
fitting an elliptic equation of the edge of the molten pool and a linear equation of the edge of the fitted weld seam according to the edge of the molten pool and the edge of the weld seam, obtaining the center of the molten pool according to the obtained elliptic equation, obtaining a vertical point from the center of the molten pool to the central line of the weld seam, and obtaining a coordinate of the center of the molten pool and a coordinate of the vertical point;
the central point of the ellipse equation is the center of a molten pool, and the linear equation is the central line of a welding seam;
respectively converting the coordinate of the center of the molten pool and the coordinate of the vertical point into world coordinates according to the conversion relation matrix of the pixel coordinate system and the world coordinate system, respectively obtaining the world coordinate of the center of the molten pool and the world coordinate of the vertical point, and obtaining the deviation between the center of the molten pool and the vertical point according to the world coordinate of the center of the molten pool and the world coordinate of the vertical point;
and controlling the industrial robot (1) to correct the deviation through the PC (11) according to the deviation.
4. The weld tracking method according to claim 3, wherein the conversion relation matrix of the pixel coordinate system and the world coordinate system is as follows:
Figure 500307DEST_PATH_IMAGE005
wherein,
Figure 221138DEST_PATH_IMAGE006
respectively in the world coordinate systemXA shaft,YShaft andZthe coordinates on the axis of the lens are,Ran outer rotation matrix is represented that is,Ta translation matrix is represented that represents the translation of the image,frepresenting the focal length of the HDR camera (4),dxanddyrespectively representing the actual distances of a pixel point in the HDR welding area image on an x axis and a y axis in a pixel coordinate system;
Figure 275288DEST_PATH_IMAGE007
and
Figure 372557DEST_PATH_IMAGE008
respectively, the origin of the HDR welding region image is at the center of the pixel coordinate systemxShaft andyan offset on the shaft;
Figure 80750DEST_PATH_IMAGE009
respectively in the pixel coordinate systemxShaft andycoordinates on the axis.
5. The weld joint tracking method according to claim 3, wherein the acquiring gradient information of the region of interest by using a Sobel operator comprises:
carrying out convolution operation on the convolution factor of the Sobel operator and the region of interest to obtainxGradient of direction
Figure 972483DEST_PATH_IMAGE010
Andygradient of direction
Figure 546552DEST_PATH_IMAGE011
The convolution factor of the Sobel operator is:
Figure 322879DEST_PATH_IMAGE012
according toxGradient of direction
Figure 744633DEST_PATH_IMAGE013
Andygradient of direction
Figure 698944DEST_PATH_IMAGE014
Acquiring gradient information of the region of interest:
Figure 838939DEST_PATH_IMAGE015
Figure 418956DEST_PATH_IMAGE016
wherein,
Figure 960796DEST_PATH_IMAGE017
the magnitude of the gradient is indicated by the scale,
Figure 584544DEST_PATH_IMAGE018
the gradient direction is indicated.
6. The weld joint tracking method according to claim 5, wherein the wavelet features in the wavelet feature matrix are screened by adopting a K-means clustering algorithm to obtainNFeatures of wavelet with maximum discrimination, and passingNThe minimum wave position and the gradient information with the maximum identification degree acquire the welding seam identification characteristics, and the method comprises the following steps:
combining gradient information, taking the gradient size and gradient direction corresponding to each pixel in a wavelet feature matrix and a first wavelet feature as feature spaces, and clustering the feature spaces corresponding to each pixel based on a K-means clustering algorithm to obtain a K-means clustering result, wherein the K value of the K-means clustering algorithm is 3, and the K-means clustering result comprises the classification of each pixel;
obtaining the intra-class difference of each class according to the K-means clustering result
Figure 211834DEST_PATH_IMAGE019
Comprises the following steps:
Figure 329963DEST_PATH_IMAGE020
wherein,iindicates the category of the K-means clustering result,ithe value ranges of {1,2,3}, {1,2,3} respectively represent the edge of the molten pool, the welding line and the background,nrepresenting categoriesiThe number of the pixel points contained in the image,
Figure 991888DEST_PATH_IMAGE021
representing categoriesiTo middlejThe wavelet characteristic value corresponding to each pixel point,j=1,2,…,n，
Figure 13635DEST_PATH_IMAGE022
representing categoriesiThe average value of wavelet characteristic values corresponding to all the pixel points in the image;
obtaining the difference between the two categories according to the K-means clustering result
Figure 128221DEST_PATH_IMAGE023
Comprises the following steps:
Figure 50041DEST_PATH_IMAGE024
wherein,
Figure 566473DEST_PATH_IMAGE025
indicates the category of the K-means clustering result,
Figure 532024DEST_PATH_IMAGE026
the value range of (1), (2), (3),
Figure 868327DEST_PATH_IMAGE027
is shown andithe different categories of the content are,
Figure 328258DEST_PATH_IMAGE028
representing categories
Figure 699197DEST_PATH_IMAGE026
The average value of wavelet characteristic values corresponding to all the pixel points in the image;
according to the difference in class
Figure 133851DEST_PATH_IMAGE029
And the difference between the two categories
Figure 832817DEST_PATH_IMAGE030
Obtaining the identification degree of the first wavelet featureCSComprises the following steps:
Figure 955494DEST_PATH_IMAGE031
wherein,
Figure 305573DEST_PATH_IMAGE032
representing categories
Figure 488292DEST_PATH_IMAGE033
Within class differences of (3);
based on the steps, traversing corresponding pixels
Figure 736871DEST_PATH_IMAGE034
A wavelet feature to obtain
Figure 286408DEST_PATH_IMAGE034
Individual identification degree;
according to the identification degree of the wavelet characteristics, the corresponding pixel point of each pixel point is obtainedNThe wavelet feature with the maximum identification degree and willNAnd forming a welding seam identification characteristic by the wavelet characteristic with the maximum identification degree and the gradient information.
7. The weld tracking method according to claim 3, wherein the support vector machine employs a Gaussian kernel function.
8. The weld tracking method according to claim 3, wherein the fitting of an elliptic equation of the edge of the weld pool and a linear equation of the edge of the weld pool based on the edge of the weld pool and the edge of the weld pool, and the obtaining of the center of the weld pool based on the obtained elliptic equations, the obtaining of the vertical point from the center of the weld pool to the center line of the weld pool, and the obtaining of the coordinates of the center of the weld pool and the coordinates of the vertical point comprise:
the initial ellipse equation is constructed as follows:
Figure 366359DEST_PATH_IMAGE035
wherein,
Figure 392084DEST_PATH_IMAGE036
the coordinates of the pixel coordinate system are represented,a、b、c、danderespectively representing a first coefficient, a second coefficient, a third coefficient, a fourth coefficient and a fifth coefficient;
random selection in the edge of the molten poolMIs spotted according toMThe coordinates of the points are used for solving the coefficients of the initial elliptic equation to obtain an iterative elliptic equation,Mis greater than or equal to 5, or,Many three of the points are not collinear, and the iterative ellipse equation is:
Figure 314910DEST_PATH_IMAGE037
wherein,mthe number of iterations is indicated and,
Figure 44968DEST_PATH_IMAGE038
、
Figure 854792DEST_PATH_IMAGE039
、
Figure 644894DEST_PATH_IMAGE040
、
Figure 556480DEST_PATH_IMAGE041
and
Figure 824651DEST_PATH_IMAGE042
respectively representmOf a minor iterationa、b、c、dAndethe solution value of (2);
traversing each point in the edge of the molten pool to obtain the deviation between each point in the edge of the molten pool and the iterative elliptic equation
Figure 754560DEST_PATH_IMAGE043
Comprises the following steps:
Figure 449984DEST_PATH_IMAGE044
wherein,
Figure 81822DEST_PATH_IMAGE045
the intermediate parameter is represented by a value representing,
Figure 419263DEST_PATH_IMAGE046
indicating the edge of the molten poolnThe coordinates of the points are such that,n=1,2,…,L，Lrepresenting the total number of the middle points of the edge of the molten pool;
make the number of iterationsm=1, repeating the above steps, and repeating the number of iterations in each repeating processmIs incremented by one until the number of iterationsmWhen a set threshold value is reached, selecting an iterative elliptical equation with the largest number of interior points as an elliptical equation of the edge of the molten pool, wherein the interior points refer to deviation
Figure 469258DEST_PATH_IMAGE047
A point less than 5;
according to an elliptic equation, the center of the molten pool is obtained as follows:
Figure 335583DEST_PATH_IMAGE048
wherein,
Figure 224692DEST_PATH_IMAGE049
coordinates representing the center of the molten pool;
the linear equation of the welding seam edge is fitted by adopting a least square method as follows:
Figure 834665DEST_PATH_IMAGE050
wherein,gwhich represents the sixth coefficient of the first coefficient,hwhich represents the seventh coefficient of the signal,
Figure 739167DEST_PATH_IMAGE051
indicating the second in the weld edgeqThe coordinates of the points are such that,q=1,2,…,Q，Qindicates the total number of midpoints of the edges of the weld,
Figure 776393DEST_PATH_IMAGE052
representing the mean value of coordinates of all points in the edge of the weld;
obtaining the coordinate of the vertical point from the center of the molten pool to the central line of the welding seam
Figure 648403DEST_PATH_IMAGE053
Comprises the following steps:
Figure 62067DEST_PATH_IMAGE054
。
9. the weld tracking method according to claim 3, wherein the converting the coordinate of the center of the molten pool and the coordinate of the vertical point into world coordinates according to a conversion relation matrix of a pixel coordinate system and a world coordinate system, respectively obtaining the world coordinate of the center of the molten pool and the world coordinate of the vertical point, respectively, and obtaining the deviation between the center of the molten pool and the vertical point according to the world coordinate of the center of the molten pool and the world coordinate of the vertical point comprises:
according to the pixel coordinate system and world coordinatesA conversion relation matrix of a standard system, which converts the coordinate of the center of the molten pool and the coordinate of the vertical point into world coordinates to respectively obtain the world coordinates of the center of the molten pool and the world coordinates of the vertical point as
Figure 821075DEST_PATH_IMAGE055
And
Figure 294782DEST_PATH_IMAGE056
；
acquiring the deviation between the center of the molten pool and the vertical point according to the world coordinates of the center of the molten pool and the world coordinates of the vertical point:
Figure 889973DEST_PATH_IMAGE057
wherein,
Figure 841749DEST_PATH_IMAGE058
between the centre of the bath and the vertical pointXThe deviation of the axis of rotation is such that,
Figure 720843DEST_PATH_IMAGE059
between the centre of the bath and the vertical pointYThe deflection of the shaft.
10. The weld seam tracking method according to claim 3, wherein the controlling the industrial robot (1) by the PC (11) to perform deviation correction according to the deviation comprises: and filtering the deviation through a Kalman filter to obtain a final welding deviation, and controlling the industrial robot (1) to correct the deviation through the PC (11) according to the final welding deviation.
CN202111273178.XA 2021-10-29 2021-10-29 HDR image-based molten pool center identification system and weld joint tracking method Active CN113828892B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111273178.XA CN113828892B (en) 2021-10-29 2021-10-29 HDR image-based molten pool center identification system and weld joint tracking method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111273178.XA CN113828892B (en) 2021-10-29 2021-10-29 HDR image-based molten pool center identification system and weld joint tracking method

Publications (2)

Publication Number Publication Date
CN113828892A CN113828892A (en) 2021-12-24
CN113828892B true CN113828892B (en) 2022-05-10

Family

ID=78966637

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111273178.XA Active CN113828892B (en) 2021-10-29 2021-10-29 HDR image-based molten pool center identification system and weld joint tracking method

Country Status (1)

Country Link
CN (1) CN113828892B (en)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114932292B (en) * 2022-05-27 2023-09-26 华南理工大学 Narrow-gap passive vision weld joint tracking method and system
CN117054334A (en) * 2023-07-21 2023-11-14 常州超音速智能装备有限公司 Top cover welding defect quality detection method and system based on multiple cameras
CN117428297B (en) * 2023-09-01 2024-04-16 广东福维德焊接技术研究院有限公司 K-TIG welding gun short circuit, full-automatic real-time detection device for welding seam and control system

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2014014855A (en) * 2012-07-11 2014-01-30 Suzuki Motor Corp Image processing method and image processing system
CN106251332A (en) * 2016-07-17 2016-12-21 西安电子科技大学 SAR image airport target detection method based on edge feature
CN107590785A (en) * 2017-09-06 2018-01-16 吉林大学 A kind of Brillouin spectrum image-recognizing method based on sobel operators
CN108637435A (en) * 2018-05-16 2018-10-12 华南理工大学 A kind of three-dimensional seam tracking system and method for view-based access control model and arc voltage sensing
CN109978874A (en) * 2019-04-02 2019-07-05 湖南大学 A kind of rail surface defects vision inspection apparatus and recognition methods
CN111127402A (en) * 2019-12-06 2020-05-08 东莞理工学院 Visual detection method for welding quality of robot
CN111283303A (en) * 2020-04-02 2020-06-16 广东福维德焊接股份有限公司 HDR image-based molten pool center identification system and method
CN111570975A (en) * 2020-05-12 2020-08-25 湖北文理学院 Welding deviation measuring method and device based on synchronous drawing and automatic welding system

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2014014855A (en) * 2012-07-11 2014-01-30 Suzuki Motor Corp Image processing method and image processing system
CN106251332A (en) * 2016-07-17 2016-12-21 西安电子科技大学 SAR image airport target detection method based on edge feature
CN107590785A (en) * 2017-09-06 2018-01-16 吉林大学 A kind of Brillouin spectrum image-recognizing method based on sobel operators
CN108637435A (en) * 2018-05-16 2018-10-12 华南理工大学 A kind of three-dimensional seam tracking system and method for view-based access control model and arc voltage sensing
CN109978874A (en) * 2019-04-02 2019-07-05 湖南大学 A kind of rail surface defects vision inspection apparatus and recognition methods
CN111127402A (en) * 2019-12-06 2020-05-08 东莞理工学院 Visual detection method for welding quality of robot
CN111283303A (en) * 2020-04-02 2020-06-16 广东福维德焊接股份有限公司 HDR image-based molten pool center identification system and method
CN111570975A (en) * 2020-05-12 2020-08-25 湖北文理学院 Welding deviation measuring method and device based on synchronous drawing and automatic welding system

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
K-TIG焊接焊缝偏差与熔透状态识别的研究;朱焘;《工程科技Ⅰ辑》;20200115;全文 *
K-TIG焊接窄间隙焊缝视觉跟踪系统研究;刘炜聪;《工程科技Ⅰ辑》;20200115;全文 *
基于图像传感技术的焊缝定位检测系统的研究;李松阳;《工程科技Ⅰ辑》;20180815;全文 *

Also Published As

Publication number Publication date
CN113828892A (en) 2021-12-24

Similar Documents

Publication Publication Date Title
CN113828892B (en) HDR image-based molten pool center identification system and weld joint tracking method
CN110539109B (en) Robot automatic welding system and method based on single-binocular vision
CN106903426B (en) A kind of laser welding localization method based on machine vision
Bdiwi et al. Autonomous disassembly of electric vehicle motors based on robot cognition
CN111604909A (en) Visual system of four-axis industrial stacking robot
CN112529858A (en) Welding seam image processing method based on machine vision
CN114140439B (en) Laser welding seam characteristic point identification method and device based on deep learning
CN110246127A (en) Workpiece identification and localization method and system, sorting system based on depth camera
CN114905507A (en) Welding robot precision control method based on environment vision analysis
CN108416809A (en) A kind of steel drum threaded cap method for recognizing position and attitude based on machine vision
CN108907526A (en) A kind of weld image characteristic recognition method with high robust
CN113146172A (en) Multi-vision-based detection and assembly system and method
CN112561886A (en) Automatic workpiece sorting method and system based on machine vision
CN115609591B (en) Visual positioning method and system based on 2D Marker and compound robot
CN111739006B (en) Elliptical image detection algorithm and system based on enclosed road integral
WO2019059343A1 (en) Workpiece information processing device and recognition method of workpiece
CN112381783B (en) Weld track extraction method based on red line laser
US20230150142A1 (en) Device and method for training a machine learning model for generating descriptor images for images of objects
CN115861780B (en) Robot arm detection grabbing method based on YOLO-GGCNN
CN113369761A (en) Method and system for guiding robot welding seam positioning based on vision
CN103020638B (en) One is carried out welding bead based on chromatic information and is known method for distinguishing
CN117037062A (en) Target object grabbing method, system, electronic equipment and storage medium
CN112588621A (en) Agricultural product sorting method and system based on visual servo
CN114926531A (en) Binocular vision based method and system for autonomously positioning welding line of workpiece under large visual field
CN113843797A (en) Automatic dismounting method for part hexagon bolt in non-structural environment based on monocular and binocular mixed vision

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant