Vision-based teachless robot compliance control system and method

Abstract

The present application relates to the field of automatic welding equipment, and is especially designed to be a vision-based teaching-free robot compliance control system and method.The present application adopts the following technical scheme: including a weld positioning module, a weld identification module, a welding gun posture planning module, a transition path planning module, an execution control module, and a general control module connected with the weld positioning module, the weld identification module, the welding gun posture planning module, the transition path planning module, and the execution control module.The present application has the beneficial effect that through the mode of dragging positioning weld + automatic weld identification, the user only needs to roughly position the weld position, and the system can automatically complete weld extraction and path generation, shortening the programming practice to the minute level, effectively applicable to the welding scene of small-batch multi-variety products, and can realize efficient and high-quality customized welding and support flexible production.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of automatic welding equipment, in particular to a vision-based teachless robot compliant control system and method. BACKGROUND

[0002] As the core process of manufacturing industry, welding accounts for more than 40% of industrial robot applications. With the growing demand for multi-variety and small-batch production, traditional automated welding faces two major contradictions: (1) rigid automation, suitable for mass production, but unable to adapt to workpiece size fluctuations or frequent production changes; (2) labor-dependent, small and medium batch scenarios still rely on workers holding welding guns or teaching programming, which is low in efficiency (single programming takes 30-120 minutes) and unstable in quality (the pass rate is usually 85-93%). How to achieve efficient and high-quality welding without sacrificing flexibility has become a common problem in the fields of equipment manufacturing, aerospace, etc. Currently, the main technical routes to solve the above problems are as follows: (1) vision-guided welding, which scans the weld through laser / structured light, but requires pre-modeling or strict positioning; (2) drag teaching, which uses the force control function of collaborative robots to record the path, but lacks process automation (welding gun posture and parameters still need to be set manually), and the transition path also needs to be set manually. As can be seen, the existing technology has the contradiction of relying on precise work (sacrificing flexibility) or only achieving partial automation, and has not formed a "perception-decision-execution" closed loop. SUMMARY

[0003] The purpose of the present application is to provide a vision-based teachless robot compliant control system and method, specifically to provide a compliant control system and method that can achieve efficient and high-quality customized automatic welding.

[0004] To achieve the above purpose, the present application adopts the following technical scheme: a vision-based teachless robot compliant control system, comprising a weld positioning module, a weld recognition module, a welding gun posture planning module, a transition path planning module, an execution control module, and a general control module connected to the weld positioning module, the weld recognition module, the welding gun posture planning module, the transition path planning module, and the execution control module.

[0005] The weld positioning module includes a torque sensor and a trajectory recording unit. The torque sensor is arranged on the motor of each joint of the robot used to drive the welding gun to move, for detecting the force received by each joint of the robot when the operator drags the welding gun, so that the robot can follow the operator to drag the welding gun to move. The trajectory recording unit is connected to the controller of the robot for recording the position and attitude data of the end effector of the robot, thereby preliminarily determining the positions of several welds on the workpiece.

[0006] The weld recognition module includes a 3D vision sensor and point cloud processing software. The 3D vision sensor is used to collect 3D point cloud data of several welds on the workpiece, and the point cloud processing software is used to process the 3D point cloud data to obtain a three-dimensional contour image of the weld.

[0007] The welding gun posture analysis module is used to determine the optimal travel path and posture of the welding gun when welding the weld based on the three-dimensional contour image of the weld, so as to make the welding gun perpendicular to the weld and avoid interference and collision between the robot and the welding gun and the workpiece during the welding process.

[0008] The transition path planning module is used to plan the transition path when the welding gun transitions between different welds to avoid collision between the robot, the welding gun and the workpiece.

[0009] The execution control module is used to control the robot to perform welding operations.

[0010] Specifically, it also includes an AI process knowledge base, which is used to provide the optimal welding current, voltage and welding speed according to the three-dimensional contour image of the weld.

[0011] Specifically, it also includes a simulation verification module, which is used to simulate the welding path according to the three-dimensional contour image of the weld to verify the feasibility of the welding path.

[0012] A method for welding using the above-mentioned vision-based teaching-free robot compliant control system includes the following steps:

[0013] S1. The operator drags the welding gun connected to the robot's end effector to move at the weld position. The weld positioning module drives the robot to perform auxiliary movement according to the direction of the force applied to the robot by the operator when dragging the welding gun, and records the position and posture of the robot's end effector to preliminarily determine the positions of several welds on the workpiece.

[0014] S2. The weld recognition module performs 3D scanning on the welds according to the positions of the welds on the workpiece preliminarily determined in step S1 to collect high-precision point cloud data of the welds, and establishes a three-dimensional contour of the welds after processing the point cloud data.

[0015] S3. The welding gun posture analysis module determines the specific welding path and the posture of the welding gun according to the three-dimensional contour of the weld, and determines the welding parameters of the welding gun according to the three-dimensional contour of the weld.

[0016] S4, the transition path planning module plans the transition path of the welding gun when welding different welds. The transition path planning module takes the initial point of the transition as the starting point, which is recorded as JPos b , with the welding starting point of the target weld as the end point, recorded as JPose , and perform transition path planning, where the starting point, end point, and each point on the transition path include the joint positions of the robot's six axes; the specific method for planning the transition path is as follows:

[0017] S41. First, preset the robot model and the point cloud model of the workpiece to be welded in the scene.

[0018] S42. Define the starting point of the transition path JPos b and the end point JPos e .

[0019] S43. Define the minimum search distance as M, and define the result storage object of all points on the transition path as JPoslist.

[0020] S44. Start two threads and perform parallel searches from the starting point to the end point and from the end point to the starting point respectively, until the distance between the search points in the two directions is less than M, and store the search points in JPoslist. The final output JPoslist is the transition path that completes the planning.

[0021] S5. The execution control module controls and drives the robot and the welding gun according to the specific welding path of the welding gun, the posture of the welding gun, and the welding parameters of the welding gun obtained in step S3, together with the planned transition path output in step S44, to complete the welding process.

[0022] Specifically, in step S44, when performing parallel searches in two directions, from the starting point to the end point and from the end point to the starting point, the following method is specifically used:

[0023] S441, define the search direction from the starting point to the end point as direction one, and define the current search point of direction one as JPos c1 , JPos c1 The initial value is equal to JPos b ; Define the search direction from the end point to the starting point as direction 2, and define the current search point of direction 2 as JPos c2 , JPos c2 The initial value is equal to JPos e .

[0024] S442, define the search target point in direction 1 as JPos t1 , JPos t1 Equal to JPos c2 , and follow JPos c2 Real-time changes; define the search target point of direction 2 as JPos t2 , JPos t2 Equal to JPos c1 , and follow JPosc1 Real-time changes.

[0025] S443: Construct a search direction based on the current search point and the search target point of direction 1. ; According to the current search point and search target point of direction 2, build the search direction ; , .

[0026] S444, respectively use the current search point JPos of direction one and direction two c1 JPos c2 As a starting point, and For the direction, construct new search points JPos for direction one and direction two n1 JPos n2 ;

[0027] JPos n1 =JPos c1 + ;

[0028] JPos n2 =JPos c2 + ; Wherein M is the minimum search distance in step S43.

[0029] S445. Utilize the robot model and the point cloud model of the workpiece to be welded preset in the scene in step S41, adjust the state of the robot according to the new search points of direction one and direction two constructed in step S444, and determine whether the robot collides with the point cloud model of the workpiece to be welded. If no collision occurs, jump to step S447; if a collision occurs, jump to step S446.

[0030] S446. When a collision occurs, the search direction is adjusted according to the joint that collides with the workpiece, a new search direction is constructed, and the process jumps to step S444 to reconstruct a new search point.

[0031] S447, when no collision occurs, calculate the midpoint between the current search point and the new search point, i.e. JPos c1 to JPos n1 The midpoint of JPos z1 =(JPos c1 +JPos n1 ) / 2 or JPos c2 to JPos n2 The midpoint of JPos z2 =(JPos c2 +JPos n2) / 2, and determine whether the robot collides with the workpiece point cloud model at the midpoint. If no collision occurs, the new search point is stored in JPoslist and the process jumps to step S448, and the new search point is used as the updated current search point, that is, the JPos is updated. c1 The value of JPos n1 Or update JPos c2 The value of JPos n2 If a collision occurs, jump to step S446.

[0032] S448, calculate the current search point JPos in the two search directions c1 and JPos c2 distance If dis>M, jump to step S443; if dis≤M, stop searching and output JPoslist.

[0033] Specifically, in step S446, when adjusting the search direction, the following formula is used for adjustment:

[0034] ;

[0035] ;

[0036] Among them, α is the random perturbation coefficient, α=0.1~0.3, is a unit random vector, β is the target direction guidance coefficient, β=0.5~0.8, is the bias vector based on the target direction, .

[0037] in The setting adopts the following method: First, select the adjustment strategy according to the axis where the robot collides with the workpiece point cloud model: define the robot's one to six axes as J1 to J6. When any axis of the robot J1 to J3 collides and no axis of J4 to J6 collides, the spindle priority strategy is adopted; when any axis of J4 to J6 collides and no axis of J1 to J3 collides, the wrist fine-tuning strategy is adopted; when any axis of J1 to J3 collides and any axis of J4 to J6 collides, the full-axis coordination strategy is adopted; different adjustment strategies are selected for different 0:

[0038] ;

[0039] .

[0040] Specifically, for the JPoslist output in step S448, a post-optimization step is also performed. The specific optimization process is as follows: Note that JPoslist = [JPos1, JPos2, ... JPos n ], each point in JPoslist is all the new search points stored in step S447, arranged in order from the starting point to the end point, where the straight line path between adjacent points has been determined not to collide with the workpiece. During optimization, all points are traversed in order to determine whether the straight line path of non-adjacent points collides with the workpiece. If the straight line path of non-adjacent points does not collide with the workpiece, all points between the non-adjacent points are deleted to reduce the number of points in JPoslist.

[0041] Specifically, in step S3, when determining the welding parameters of the welding gun according to the three-dimensional profile of the weld, the AI ​​process knowledge base is used to select and determine the welding parameters.

[0042] Specifically, in step S5, before the execution control module controls and drives the robot and welding gun, the simulation verification module is first used to simulate the welding path according to the three-dimensional contour image of the weld to verify the feasibility of the welding path. The welding process is performed only after confirmation.

[0043] The beneficial effects of the present invention are: through the method of weld positioning + automatic weld identification, the user only needs to roughly locate the weld position, and the system can automatically complete weld extraction and path generation, shortening the programming practice to minutes, and is effectively suitable for welding scenarios of small batches and multiple varieties of products. It can achieve efficient and high-quality customized welding and support flexible production; at the same time, the transition path is effectively planned to improve the smoothness of the welding path and the welding quality; an AI process knowledge base is also set up to set the posture and parameters of the welding gun according to the weld morphology, which is convenient for operators to select parameters and improve programming efficiency and the stability of welding quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Attachment Figure 1 4 is a system block diagram of a vision-based teaching-free robot compliance control system in an embodiment. DETAILED DESCRIPTION

[0045] Example 1, with reference to Figure 1 A vision-based teaching-free robot compliant control system includes a weld positioning module, a weld recognition module, a welding gun posture planning module, a transition path planning module, an execution control module, and a master control module connected to the weld positioning module, weld recognition module, welding gun posture planning module, transition path planning module, and execution control module.

[0046] The weld seam positioning module includes a torque sensor and a trajectory recording unit. The torque sensor is installed on the motors at each joint of the robot that drives the welding gun. It detects the force exerted on each joint when the operator drags the welding gun, enabling the robot to move in accordance with the operator's movements. The trajectory recording unit is connected to the robot's controller to record the position and posture data of the robot's end effector, thereby preliminarily determining the locations of several welds on the workpiece. The weld seam positioning module is used to perform preliminary weld seam positioning, facilitating weld seam scanning and identification by the weld seam recognition module.

[0047] The weld seam recognition module includes a 3D vision sensor and point cloud processing software. The 3D vision sensor is used to collect 3D point cloud data of several welds on the workpiece, and the point cloud processing software is used to process the 3D point cloud data to obtain a 3D contour image of the weld seam. The 3D vision sensor can be a structured light camera or a laser scanner.

[0048] The welding gun posture analysis module is used to determine the optimal travel path and posture of the welding gun when welding the weld based on the three-dimensional contour image of the weld, so as to make the welding gun perpendicular to the weld and avoid interference and collision between the robot and the welding gun and the workpiece during the welding process.

[0049] The transition path planning module is used to plan the transition path when the welding gun transitions between different welds to avoid collision between the robot, the welding gun and the workpiece.

[0050] The execution control module is used to control the robot's welding operations. In this embodiment, the execution control module is composed of a Wasu Type III controller, a welding power interface, and an anomaly monitoring unit. The Wasu Type III controller precisely controls the robot's motion with a 4ms cycle; the welding power interface enables synchronous communication with the welding machine, ensuring accurate execution of welding parameters; and the anomaly monitoring unit triggers protection mechanisms by real-time monitoring of signals such as arc voltage.

[0051] The master control module uses HSRStudio master control software, which is responsible for coordinating the operating sequence and data interaction of each functional module to ensure seamless connection of the entire process from dragging and positioning welds to welding execution. The human-computer interaction interface it provides provides welding status display and parameter adjustment functions, which facilitate operator monitoring and intervention.

[0052] Specifically, it also includes an AI process knowledge base, which is used to provide the optimal welding current, voltage, and welding speed based on the three-dimensional contour image of the weld. Specifically, the AI ​​process knowledge base consists of a welding parameter database and an intelligent matching algorithm. The welding parameter database stores process parameter combinations for different materials, plate thicknesses, and weld types; the intelligent matching algorithm (combining a rule engine and machine learning technology) automatically recommends optimal welding current, voltage, speed and other parameters based on real-time identification of weld characteristics. The function of this module is to achieve intelligent matching of welding process parameters, reduce dependence on operator experience, and continuously optimize parameter recommendation strategies through online learning functions to improve the stability of welding quality.

[0053] Specifically, it also includes a simulation verification module, which is used to simulate the welding path based on the three-dimensional contour image of the weld to verify the feasibility of the welding path. The simulation verification module includes a digital twin environment, a welding process simulator, and a collision detector. The digital twin environment simulates the robot's kinematic and dynamic characteristics through a physics engine; the welding process simulator predicts the molten pool morphology and thermal deformation effects during the welding process; and the collision detector previews the robot's motion trajectory in a virtual environment to identify potential interference risks. The role of this module is to fully verify the generated program before actual welding, ensure the feasibility of the path, and optimize the robot code, thereby reducing on-site debugging time and material waste.

[0054] A method for welding using the above-mentioned vision-based teaching-free robot compliant control system includes the following steps:

[0055] S1. The operator drags the welding gun connected to the robot's end effector to move at the weld position. The weld positioning module drives the robot to perform auxiliary movement according to the direction of the force applied to the robot by the operator when dragging the welding gun, and records the position and posture of the robot's end effector, thereby preliminarily determining the positions of several welds on the workpiece; when the operator drags the welding gun, he drags the welding gun over all welds that need to be welded in sequence according to the welding order, and records the positions of all welds, thereby preliminarily determining the positions of several welds on the workpiece.

[0056] S2: The weld seam recognition module performs 3D scanning on the weld seams based on their locations initially determined in step S1, collects high-precision point cloud data, and processes the point cloud data to create a 3D profile of the weld seams. The 3D profile of the weld seams includes geometric parameters such as the weld seam width, depth, and groove angle.

[0057] S3, the welding gun posture analysis module, determines the specific welding path and posture of the welding gun based on the three-dimensional weld contour. It also determines the welding parameters based on the three-dimensional weld contour. Because the welds on a workpiece are not continuous but consist of multiple independent welds, the robot needs to transfer the welding gun between different welds when controlling the welding gun. This transfer process is called a transition path.

[0058] S4, the transition path planning module plans the transition path of the welding gun when welding different welds. The transition path planning module takes the initial point of the transition as the starting point, which is recorded as JPos b , with the welding starting point of the target weld as the end point, recorded as JPos e , and perform transition path planning, where the starting point, end point, and each point on the transition path include the joint positions of the robot's six axes; the specific method for planning the transition path is as follows:

[0059] S41. First, preset the robot model and the point cloud model of the workpiece to be welded in the scene.

[0060] S42. Define the starting point of the transition path JPos b and the end point JPos e .

[0061] S43. Define the minimum search distance as M, and define the result storage object of all points in the transition path as JPoslist; where M is obtained by multiplying the longest side of the bounding box of the workpiece point cloud model preset in step S41 by a fixed percentage. The percentage can be adjusted according to actual needs, and the default value of the percentage is 10%.

[0062] S44. Start two threads and perform parallel searches from the starting point to the end point and from the end point to the starting point respectively, until the distance between the search points in the two directions is less than M, and store the search points in JPoslist. The final output JPoslist is the transition path that completes the planning.

[0063] S5. The execution control module controls and drives the robot and the welding gun according to the specific welding path of the welding gun, the posture of the welding gun, and the welding parameters of the welding gun obtained in step S3, together with the planned transition path output in step S44, to complete the welding process.

[0064] Specifically, in step S44, when performing parallel searches in two directions, from the starting point to the end point and from the end point to the starting point, the following method is specifically used:

[0065] S441, define the search direction from the starting point to the end point as direction one, and define the current search point of direction one as JPos c1, JPos c1 The initial value is equal to JPos b ; Define the search direction from the end point to the starting point as direction 2, and define the current search point of direction 2 as JPos c2 , JPos c2 The initial value is equal to JPos e .

[0066] S442, define the search target point in direction 1 as JPos t1 , JPos t1 Equal to JPos c2 , and follow JPos c2 Real-time changes; define the search target point of direction 2 as JPos t2 , JPos t2 Equal to JPos c1 , and follow JPos c1 Real-time changes.

[0067] S443: Construct a search direction based on the current search point and the search target point of direction 1. ; According to the current search point and search target point of direction 2, build the search direction ; , .

[0068] S444, respectively use the current search point JPos of direction one and direction two c1 , JPos c2 As a starting point, and For the direction, construct new search points JPos for direction one and direction two n1 JPos n2 ;

[0069] JPos n1 =JPos c1 + ;

[0070] JPos n2 =JPos c2 + ; Wherein M is the minimum search distance in step S43.

[0071] S445. Utilize the robot model and the point cloud model of the workpiece to be welded preset in the scene in step S41, adjust the state of the robot according to the new search points of direction one and direction two constructed in step S444, and determine whether the robot collides with the point cloud model of the workpiece to be welded. If no collision occurs, jump to step S447; if a collision occurs, jump to step S446.

[0072] S446. When a collision occurs, the search direction is adjusted according to the joint that collides with the workpiece, a new search direction is constructed, and the process jumps to step S444 to reconstruct a new search point.

[0073] S447, when no collision occurs, calculate the midpoint between the current search point and the new search point, i.e. JPos c1 to JPos n1 The midpoint of JPos z1 =(JPos c1 +JPos n1 ) / 2 or JPos c2 to JPos n2 The midpoint of JPos z2 =(JPos c2 +JPos n2 ) / 2, and determine whether the robot collides with the workpiece point cloud model at the midpoint. If no collision occurs, the new search point is stored in JPoslist and the process jumps to step S448, and the new search point is used as the updated current search point, that is, the JPos is updated. c1 The value of JPos n1 Or update JPos c2 The value of JPos n2 If a collision occurs, the process jumps to step S446. It should be noted that JPoslist initially includes the starting point JPos b and the end point JPos e Whenever a new search point is stored in JPoslist, it is arranged in order from the starting point to the end point and recorded in order as JPos1, JPos2, ... JPos n .

[0074] S448, calculate the current search point JPos in the two search directions c1 and JPos c2 distance If dis>M, jump to step S443; if dis≤M, stop searching and output JPoslist.

[0075] Specifically, in step S446, when adjusting the search direction, the following formula is used for adjustment:

[0076] ;

[0077] ;

[0078] Among them, α is the random perturbation coefficient, α=0.1~0.3, is a unit random vector, β is the target direction guidance coefficient, β=0.5~0.8, is the bias vector based on the target direction, When adjusting the search direction, a unit random vector and a bias vector based on the target direction are added to the original search direction. The unit random vector is used to make avoidance adjustments based on the axis where the robot collides, and the bias vector based on the target direction is used to avoid large deviations in the search direction. Both vectors use a certain weighting coefficient to adjust the degree of their effect on the search direction.

[0079] in The setting adopts the following method: first, the adjustment strategy is selected according to the axis where the robot collides with the workpiece point cloud model: the robot's one to six axes are defined as J1~J6, where the definition of one to six axes is the general definition of each axis of a conventional six-axis industrial robot. Specifically, the robot's one axis is the rotary axis connected to the base, the second axis is the front and back swing axis of the main arm, the third axis is the front and back swing axis of the auxiliary arm, the fourth axis is the rotation axis of the upper arm, the fifth axis is the up and down swing axis of the wrist, and the sixth axis is the fine-tuning rotation axis of the wrist; when any axis J1~J3 of the robot collides and no collision occurs between J4~J6 axes, the main axis priority strategy is adopted; when any axis J4~J6 collides and no collision occurs between J1~J3 axes, the wrist fine-tuning strategy is adopted; when any axis J1~J3 collides and any axis J4~J6 collides, the full-axis coordination strategy is adopted; different adjustment strategies are selected for different 0:

[0080] ;

[0081] .

[0082] Specifically, for the JPoslist output in step S448, a post-optimization step is also performed. The specific optimization process is as follows: Note that JPoslist = [JPos1, JPos2, ... JPos n], each point in JPoslist is all the new search points stored in step S447, arranged in order from the starting point to the end point, wherein the straight line path between adjacent points has been determined not to collide with the workpiece. During optimization, all points are traversed in order to determine whether the straight line path of non-adjacent points collides with the workpiece. If the straight line path of non-adjacent points does not collide with the workpiece, all points between non-adjacent points are deleted to reduce the number of points in JPoslist. When performing the above optimization, the logic algorithm of inner and outer loops can be used to implement it: the outer loop starts traversing from the first point of JPoslist, and enters the inner loop after determining the point of the outer loop. Taking the outer loop as the first point as an example, the traversal starts from the 1+2th point to verify whether the path from the first point to the remaining non-adjacent points collides. After the traversal is completed, the inner loop ends and returns to the outer loop, and the 2nd, 3rd, ... points of JPoslist are traversed in turn until the loop ends at the n-2th point of JPoslist. For example, JPoslist = [JPos1, JPos2, JPos3, JPos4, JPos5], where JPos1 to JPos3 do not collide, while the other non-adjacent points all collide. Then delete all points from JPos1 to JPos3, that is, delete JPos2, and the optimized JPoslist = [JPos1, JPos3, JPos4, JPos5].

[0083] Specifically, in step S3, when determining the welding parameters of the welding gun according to the three-dimensional profile of the weld, the AI ​​process knowledge base is used to select and determine the welding parameters.

[0084] Specifically, in step S5, before the execution control module controls and drives the robot and welding gun, the simulation verification module is first used to simulate the welding path according to the three-dimensional contour image of the weld to verify the feasibility of the welding path. The welding process is performed only after confirmation.

[0085] Of course, the above are only preferred embodiments of the present invention and are not intended to limit the scope of use of the present invention. Therefore, any equivalent changes based on the principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for welding using a vision-based, teaching-free robot compliant control system, characterized in that: The vision-based teaching-free robot compliance control system includes a weld positioning module, a weld recognition module, a welding gun posture planning module, a transition path planning module, an execution control module, and a master control module connected to the weld positioning module, the weld recognition module, the welding gun posture planning module, the transition path planning module, and the execution control module; The weld positioning module includes a torque sensor and a trajectory recording unit. The torque sensor is set on the motor of each joint of the robot used to drive the welding gun to move the welding, and is used to detect the force exerted on each joint of the robot when the operator drags the welding gun, so that the robot can move along with the operator dragging the welding gun. The trajectory recording unit is connected to the controller of the robot and is used to record the position and posture data of the robot end effector, so as to preliminarily determine the position of several welds on the workpiece. The weld seam recognition module includes a 3D vision sensor and point cloud processing software. The 3D vision sensor is used to collect 3D point cloud data of several weld seams on the workpiece, and the point cloud processing software is used to process the 3D point cloud data to obtain a three-dimensional contour image of the weld seam. The welding gun posture planning module is used to determine the optimal travel path and posture of the welding gun when welding the weld according to the three-dimensional contour image of the weld, so as to make the welding gun perpendicular to the weld and avoid interference and collision between the robot and the welding gun and the workpiece during the welding process; The transition path planning module is used to plan the transition path when the welding gun transitions between different welds to avoid collision between the robot, the welding gun and the workpiece; The execution control module is used to control the robot to perform welding operations; The steps include: S1. The operator drags the welding gun connected to the robot's end effector to move at the weld position. The weld positioning module drives the robot to perform auxiliary movement according to the direction of the force applied by the operator when dragging the welding gun, and records the position and posture of the robot's end effector to preliminarily determine the positions of several welds on the workpiece. S2, the weld recognition module performs 3D scanning on the welds according to the positions of the welds on the workpiece preliminarily determined in step S1 to collect high-precision point cloud data of the welds, and processes the point cloud data to establish a three-dimensional contour of the welds; S3, the welding gun posture planning module determines the specific welding path and the posture of the welding gun according to the three-dimensional contour of the weld, and determines the welding parameters of the welding gun according to the three-dimensional contour of the weld; S4, the transition path planning module plans the transition path of the welding gun when welding different welds. The transition path planning module takes the initial point of the transition as the starting point, which is recorded as JPos b , with the welding starting point of the target weld as the end point, recorded as JPos e , perform transition path planning, The starting point, the end point, and each point on the transition path include the joint positions of the robot's six axes. The following method is used to plan the transition path: S41, presetting the robot model and the point cloud model of the workpiece to be welded in the scene; S42. Define the starting point of the transition path JPos b and the end point JPos e ; S43, define the minimum search distance as M, and define the result storage object of all points of the transition path as JPoslist; S44. Start two threads and perform parallel searches from the starting point to the end point and from the end point to the starting point, respectively, until the distance between the search points in the two directions is less than M. The search points are stored in JPoslist. The final output JPoslist is the transition path that completes the planning. When performing parallel searches from the starting point to the end point and from the end point to the starting point, the following method is specifically used: S441, define the search direction from the starting point to the end point as direction one, and define the current search point of direction one as JPos c1 , JPos c1 The initial value is equal to JPos b ; Define the search direction from the end point to the starting point as direction 2, and define the current search point of direction 2 as JPos c2 , JPos c2 The initial value is equal to JPos e ; S442, define the search target point in direction 1 as JPos t1 , JPos t1 Equal to JPos c2 , and follow JPos c2 Real-time changes; define the search target point of direction 2 as JPos t2 , JPos t2 Equal to JPos c1 , and follow JPos c1 Real-time changes; S443: Construct a search direction based on the current search point and the search target point of direction 1. ; According to the current search point and search target point of direction 2, build the search direction ; , ; S444, respectively use the current search point JPos of direction one and direction two c1 , JPos c2 As a starting point, and For the direction, construct new search points JPos for direction one and direction two n1 JPos n2 ; JPos n1 =JPos c1 + ; JPos n2 =JPos c2 + ; Where M is the minimum search distance in step S43; S445: Using the robot model and the point cloud model of the workpiece to be welded preset in the scene in step S41, the robot state is adjusted according to the new search points in direction one and direction two constructed in step S444, and it is determined whether the robot collides with the point cloud model of the workpiece to be welded. If no collision occurs, the process jumps to step S447; if a collision occurs, the process jumps to step S446. S446: When a collision occurs, the search direction is adjusted according to the joint that collides with the workpiece, a new search direction is constructed, and the process jumps to step S444 to reconstruct a new search point. When adjusting the search direction, the following formula is used for adjustment: ； ； Among them, α is the random perturbation coefficient, α=0.1~0.3, is a unit random vector, β is the target direction guidance coefficient, β=0.5~0.8, is the bias vector based on the target direction, ; in The setting adopts the following method: First, select the adjustment strategy according to the axis where the robot collides with the workpiece point cloud model: define the robot's one to six axes as J1 to J6. When any axis of the robot J1 to J3 collides and no axis of J4 to J6 collides, the spindle priority strategy is adopted; when any axis of J4 to J6 collides and no axis of J1 to J3 collides, the wrist fine-tuning strategy is adopted; when any axis of J1 to J3 collides and any axis of J4 to J6 collides, the full-axis coordination strategy is adopted; different adjustment strategies are selected for different 0: ； ； S447, when no collision occurs, calculate the midpoint between the current search point and the new search point, i.e. JPos c1 to JPos n1 The midpoint of JPos z1 =(JPos c1 +JPos n1 ) / 2 or JPos c2 to JPos n2 The midpoint of JPos z2 =(JPos c2 +JPos n2 ) / 2, and determine whether the robot collides with the workpiece point cloud model at the midpoint. If no collision occurs, the new search point is stored in JPoslist and the process jumps to step S448, and the new search point is used as the updated current search point, that is, the JPos is updated. c1 The value of JPos n1 Or update JPos c2 The value of JPos n2 If a collision occurs, jump to step S446; S448, calculate the current search point JPos in the two search directions c1 and JPos c2 distance If dis>M, jump to step S443; if dis≤M, stop searching and output JPoslist; S5. The execution control module controls and drives the robot and the welding gun according to the specific welding path of the welding gun, the posture of the welding gun, and the welding parameters of the welding gun obtained in step S3, together with the planned transition path output in step S44, to complete the welding process.

2. The method for welding using a vision-based teaching-free robot compliant control system according to claim 1, characterized in that: For the JPoslist output in step S448, a post-optimization step is also performed. The specific optimization process is as follows: JPoslist = [JPos1, JPos2, ... JPos n ], each point in JPoslist is all the new search points stored in step S447, arranged in order from the starting point to the end point, where the straight line path between adjacent points has been determined not to collide with the workpiece. During optimization, all points are traversed in order to determine whether the straight line path of non-adjacent points collides with the workpiece. If the straight line path of non-adjacent points does not collide with the workpiece, all points between the non-adjacent points are deleted to reduce the number of points in JPoslist.

3. The method for welding using a vision-based teaching-free robot compliant control system according to claim 1, characterized in that: In step S3, when determining the welding parameters of the welding gun according to the three-dimensional profile of the weld, the AI ​​process knowledge base is used to select and determine the welding parameters.

4. The method for welding using a vision-based teaching-free robot compliant control system according to claim 1, characterized in that: In step S5, before the control module controls and drives the robot and the welding gun, the simulation verification module is first used to simulate the welding path according to the three-dimensional contour image of the weld to verify the feasibility of the welding path. The welding process is performed only after confirmation.

Images