Self-adaptive visual collaborative robot welding system and method for iron tower feet

Abstract

The invention relates to the technical field of robot welding, and discloses a self-adaptive visual collaborative iron tower foot robot welding system and method.The method comprises the steps that binocular vision and laser radar composite scanning is conducted on iron tower foot angle steel, and three-dimensional coordinates of welding points are obtained; joint two-dimensional coordinate mapping conversion is conducted on the welding point three-dimensional coordinates, the robot joint two-dimensional angle is obtained, and the joint radial speed is calculated; welding energy balance calculation is conducted according to the joint radial speed, and balance control parameters are obtained; and calculating a joint compensation amount and a power regulation factor according to the balance control parameter, and generating a first welding control instruction. According to the method, a traditional six-dimensional joint space inverse kinematics solution is subjected to dimensionality reduction to a two-dimensional key motion plane of base rotation and main arm bending, the calculation complexity is greatly reduced, the problems of multiple solutions and no solutions are solved, the iron tower foot robot can adapt to angle steel of different thicknesses and change assembly gaps, and the welding quality stability is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of robot welding, and in particular to a self-adaptive visual collaborative tower leg robot welding system and method. BACKGROUND

[0002] The tower leg is a key load-bearing structure of a power transmission line. A traditional tower leg robot welding system mainly relies on a single visual sensor for welding point positioning. However, in an actual welding process, welding spatter, smoke, and a complex angle steel structure of the tower leg can seriously interfere with the visual detection accuracy. The traditional method needs to solve inverse kinematics in a six-dimensional joint space, which has high computational complexity and poor real-time performance. For a welding object such as the tower leg with specific geometric constraints, six-dimensional solving often produces multiple solutions or no solution, which affects the stability and continuity of the welding trajectory, resulting in low energy utilization and uneven heat input in the welding process, and it is difficult to form a good weld formation. SUMMARY

[0003] The present application provides a self-adaptive visual collaborative tower leg robot welding system and method. The present application reduces the dimensionality of traditional six-dimensional joint space inverse kinematics solving to a two-dimensional key motion plane of base rotation and main arm bending, greatly reduces the computational complexity, eliminates multiple solutions and no solution problems, and enables the tower leg robot to adapt to different thickness angle steels and changing assembly gaps, improving the stability of welding quality.

[0004] In a first aspect, the present application provides a self-adaptive visual collaborative tower leg robot welding method, which comprises: Performing binocular vision and laser radar composite scanning on the tower leg angle steel to obtain three-dimensional coordinates of the welding points; Converting the three-dimensional coordinates of the welding points to joint two-dimensional coordinates to obtain robot joint two-dimensional angles and calculate joint radial velocities; Performing welding energy balance calculation according to the joint radial velocities to obtain balance control parameters; Calculating joint compensation amounts and power adjustment factors according to the balance control parameters and generating first welding control instructions.

[0005] In a first implementation manner of the first aspect, the binocular vision and laser radar composite scanning on the tower leg angle steel to obtain three-dimensional coordinates of the welding points comprises: Acquiring RGB image data of the tower leg angle steel by a binocular camera, and simultaneously acquiring three-dimensional point cloud data of the tower leg angle steel by a laser radar; Performing voxel filtering and plane fitting on the three-dimensional point cloud data to extract angle steel edge line features; input the RGB image data into a deep learning network for convolution calculation and full connection layer inference, and output a welding groove type probability distribution; perform spatial position fusion based on the angle steel edge line feature and the welding groove type probability distribution to obtain a welding point three-dimensional coordinate.

[0006] In combination with the first aspect, in a second implementation manner of the first aspect of the present application, the welding point three-dimensional coordinate is mapped and converted into a robot joint two-dimensional coordinate to obtain a robot joint two-dimensional angle, and a joint radial velocity is calculated, including: In combination with the welding point three-dimensional coordinate, a motion analysis is performed on the tower leg robot to determine a motion plane of base rotation and main arm bending, and the motion plane of base rotation and main arm bending is taken as a two-dimensional mapping dimension reduction parameter; Based on the two-dimensional mapping dimension reduction parameter, an inverse tangent function calculation is performed to obtain a base rotation angle, and a cosine theorem is used to calculate a main arm joint angle, and the base rotation angle and the main arm joint angle are taken as initial joint two-dimensional mapping values of the tower leg robot; A Jacobian matrix operation is performed on the initial joint two-dimensional mapping values to obtain a target joint angle; A visual positioning deviation vector is calculated according to the target joint angle and a welding point target position, and a robot joint two-dimensional angle is generated through a PID controller; Based on the robot joint two-dimensional angle, a radial velocity calculation is performed to obtain a joint radial velocity.

[0007] In combination with the first aspect, in a third implementation manner of the first aspect of the present application, the visual positioning deviation vector is calculated according to the target joint angle and a welding point target position, and a robot joint two-dimensional angle is generated through a PID controller, including: Based on the target joint angle, an actual position coordinate of an end effector of the tower leg robot is calculated through a forward kinematics equation; A three-dimensional space difference operation is performed according to the actual position coordinate and a welding point target position to obtain a visual positioning deviation vector; The visual positioning deviation vector is input into a PID controller for proportional, integral and differential operation and calculation of a joint angle compensation amount to obtain a joint compensation control amount; Based on the joint compensation control amount, the target joint angle is superimposed and corrected to obtain a robot joint two-dimensional angle.

[0008] In combination with the first aspect, in a fourth implementation manner of the first aspect of the present application, based on the robot joint two-dimensional angle, a radial velocity calculation is performed to obtain a joint radial velocity, including: A differential calculation is performed on the robot joint two-dimensional angle to obtain an angular velocity of each joint; The joint angular velocity is multiplied by a velocity Jacobian matrix to obtain a radial motion velocity of the welding torch; A radial velocity optimization objective function including a joint rotational inertia term, an angular acceleration smoothing term and a velocity tracking term is established based on the radial motion velocity of the welding torch; A sequence quadratic programming algorithm is used to perform a Hessian matrix iteration solution on the radial velocity optimization objective function to obtain a joint radial velocity.

[0009] In a fifth implementation manner of the first aspect, the welding energy balance calculation according to the joint radial velocity to obtain a balance control parameter comprises: The joint radial velocity is divided by an arc efficiency and a welding voltage and current parameter to obtain an instantaneous input energy; The instantaneous input energy is substituted into a partial differential equation of molten pool heat conduction including a thermal diffusion coefficient and a steel thermal physical property parameter to perform a numerical solution to obtain a welding molten pool temperature field distribution; Heat transfer calculation is performed on the molten pool temperature field distribution and a convection heat transfer coefficient and an emissivity to obtain a heat dissipation loss power; The instantaneous input energy and the heat dissipation loss power are substituted into an energy balance equation to perform a coupled optimization solution to obtain the balance control parameter.

[0010] In a sixth implementation manner of the first aspect, the coupled optimization solution of the instantaneous input energy and the heat dissipation loss power into the energy balance equation to obtain the balance control parameter comprises: The instantaneous input energy and the heat dissipation loss power are subjected to a subtraction operation and an energy balance equation is established with a product of a molten pool mass and a specific heat capacity; A welding current range constraint and a radial velocity boundary constraint are set based on the energy balance equation; The welding current range constraint and the radial velocity boundary constraint are subjected to a gradient descent solution with a welding current and a radial velocity to obtain the balance control parameter.

[0011] In a seventh implementation manner of the first aspect, the calculation of a joint compensation amount and a power adjustment factor according to the balance control parameter and the generation of a first welding control instruction comprises: Difference operation is performed on the balance control parameter and a real-time motion state of a tower leg robot to obtain visual positioning deviation data, speed deviation data and energy deviation data; The visual positioning deviation data, the speed deviation data and the energy deviation data are input into a fuzzy neural network to perform fuzzy processing of a Gaussian membership function to obtain three-way fuzzy input signals; Performing a hyperbolic tangent activation operation and a weight matrix calculation based on the three-way fuzzified input signal to obtain a joint compensation amount and a power adjustment factor; The joint compensation amount and the two-dimensional angle of the robot joint are superimposed and calculated, and the power adjustment factor and the welding current are adjusted and calculated to obtain a first welding control instruction.

[0012] In combination with the first aspect, in an eighth implementation of the first aspect of the present invention, the adaptive vision-coordinated tower leg robot welding method further includes: Line structured light projection and image acquisition are performed on the weld surface during the tower foot welding process to obtain weld contour deformation data; Calculating weld cross-sectional geometric dimensions and extracting weld geometric feature data based on the weld profile deformation data; Performing quality assessment on the weld geometric feature data to obtain quality evaluation parameters; The joint compensation amount and the power adjustment factor in the first welding control instruction are adjusted according to the quality evaluation parameter through a gradient ascent algorithm to generate a second welding control instruction.

[0013] In a second aspect, the present invention provides an adaptive visual collaborative iron tower foot robot welding system, the adaptive visual collaborative iron tower foot robot welding system comprising: The composite scanning module is used to perform binocular vision and laser radar composite scanning on the tower foot angle steel to obtain the three-dimensional coordinates of the welding points; A coordinate mapping conversion module is used to convert the three-dimensional coordinates of the welding point into two-dimensional coordinates of the joint, obtain the two-dimensional angle of the robot joint, and calculate the radial velocity of the joint; a welding energy balance calculation module, configured to perform welding energy balance calculation according to the joint radial velocity to obtain balance control parameters; A generation module is used to calculate the joint compensation amount and the power adjustment factor according to the balance control parameters and generate a first welding control instruction.

[0014] The technical solution provided by this invention combines RGB image information acquired by a binocular camera with high-precision 3D point cloud data provided by a laser radar (LiDAR) to form a dual-constraint vision-geometry recognition mechanism. This effectively overcomes the vulnerability of a single vision system to interference from welding spatter and smoke, improving the recognition accuracy and anti-interference capability of the tower foot angle steel welds. The traditional six-dimensional joint space inverse kinematics solution is reduced to a two-dimensional critical motion plane of base rotation and main arm bending, significantly reducing computational complexity and eliminating multiple solutions and unsolvable problems. A coupled optimization model is established for the robot joint radial velocity and welding energy input. Motion and process parameters are simultaneously optimized using a sequential quadratic programming algorithm, addressing the low energy efficiency inherent in traditional methods, which results from independent optimization of motion control and welding process. A fuzzy neural network is employed to integrate visual feedback, motion state, and energy state information. Control parameters are dynamically adjusted through an online learning mechanism, enabling the tower foot robotic welding system to adapt to varying angle steel thicknesses and varying assembly gaps, improving weld quality stability. Laser profile scanning monitors weld geometry in real time, and a gradient ascent algorithm is employed for parameter feedback adjustment, ensuring consistent and controllable quality during the tower foot welding process.

[0015] Other features and advantages of the present invention will be described in the following description, and in part will become apparent from the description, or understood by practicing the present invention. The purposes and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description, claims and drawings.

[0016] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A schematic diagram of an embodiment of a tower leg welding method using an adaptive visual collaboration robot according to an embodiment of the present invention; Figure 2 This is a schematic diagram of an embodiment of an adaptive visual collaborative tower leg robot welding system in an embodiment of the present invention. DETAILED DESCRIPTION

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0019] The terms "including," "having," and any variations thereof, as used in the embodiments of the present invention are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or device comprising a series of steps or units is not limited to the listed steps or units, but may optionally include other steps or units not listed, or may optionally include other steps or units inherent to the process, method, product, or device.

[0020] To facilitate understanding of this embodiment, the adaptive visual collaborative tower leg robot welding method disclosed in the embodiment of the present invention is first introduced in detail. Figure 1 As shown, this method includes the following steps: 101. Perform binocular vision and laser radar composite scanning on the tower foot angle steel to obtain the three-dimensional coordinates of the welding point; Specifically, a binocular camera was used to capture high-resolution RGB images of the angle steel area at the tower foot, while a LiDAR system was activated to perform structural contour scanning to obtain three-dimensional point cloud data covering the entire angle steel area. The point cloud data contains a large amount of invalid and redundant data, so voxel filtering technology was used to sparse the data. The voxel grid side length was uniformly set to 2mm, and neighboring points were uniformly projected to the voxel center through spatial clustering to construct a sparse point cloud model with high structural fidelity. Based on the sparse point cloud model, an improved RANSAC algorithm was introduced for plane fitting calculations. During the iterative process, the minimum distance residual was used as the evaluation indicator to extract the equations of multiple planes where the main structural surface of the angle steel is located. The intersection of the two planes was extracted by calculating the angle between the normal vectors of adjacent planes as the characteristic line segment of the angle steel edge line. The RGB image is input into a deep convolutional neural network (TFWNet), which consists of multiple convolutional and fully connected layers. Through layer-by-layer feature extraction and processing with nonlinear activation functions, semantic information such as texture, brightness, and shape of the weld area in the image is extracted. The output layer generates a probability distribution of weld groove types, covering a variety of common weld structures such as butt joints, T-joints, and fillet joints. The probability of each type appearing in the current input image is described in terms of probability. The three-dimensional structural features of the angle steel edge line and the weld groove type probability distribution are aligned and semantically fused in the same spatial reference coordinate system. A feature association algorithm is used to comprehensively determine the spatial projection position of the weld near the edge line. The known weld structure type is used to determine the groove start, end, and centerline coordinates, and the three-dimensional spatial coordinates of the weld point are output.

[0021] 102. Convert the three-dimensional coordinates of the welding point to the two-dimensional coordinates of the joint to obtain the two-dimensional angle of the robot joint and calculate the radial velocity of the joint; Specifically, the kinematics of the tower foot robot structure is analyzed in combination with the three-dimensional spatial coordinates of the welding points, and it is determined that the robot mainly relies on the rotation of the base and the bending of the main arm to complete the spatial positioning operation of the welding head end during the execution of the welding operation according to the typical application working conditions. Therefore, the motion plane where the two degrees of freedom are located is determined as the parameter basis of the two-dimensional dimension reduction mapping, and the complex six-degree-of-freedom inverse kinematics problem is simplified to a two-dimensional mapping solving task. Based on the simplified model, the projection coordinates of the welding points in the XY plane are subjected to arctangent function operation, and the rotation angle of the robot base is calculated. And based on the radial distance from the welding point to the robot base, combined with the Z height coordinate, the main arm joint angle is calculated according to the triangular geometric relationship and the cosine theorem to obtain the initial two-dimensional mapping value of the joint. The initial two-dimensional angle value is input into the robot motion model, the linear differential transformation of the initial two-dimensional angle value is performed based on the constructed Jacobian matrix to obtain the sensitivity matrix of the change of the angle to the change of the pose of the end effector, and the target joint angle required to realize the target position of the end welding point under the configuration is obtained through the inverse operation of the Jacobian matrix, and the mapping relationship between the robot end welding coordinates and the joint angle is established. Considering that the positioning of the welding point in the actual operation environment will deviate due to visual error, structural shielding or coordinate drift, a visual positioning deviation vector is constructed based on the three-dimensional position difference between the target welding point and the current execution point. The visual positioning deviation vector is input into the PID controller, and a set of corrected joint two-dimensional angle values are output through proportional, integral and differential three gain adjustments. The corrected two-dimensional joint angle is taken as the input, the angular velocity of each joint is calculated according to the velocity Jacobian matrix under the current robot configuration, and then the corresponding end radial velocity is calculated by combining the two angular velocities to obtain the joint radial velocity value.

[0022] 103. Perform welding energy balance calculation according to the joint radial velocity to obtain balance control parameters; Specifically, the current joint radial velocity of the robot is taken as a dynamic variable, and the welding required current and voltage parameters, and the physical factors considering the welding efficiency are processed to calculate the heat input intensity corresponding to the unit length of the weld at the current speed, that is, the instantaneous input energy, which reflects the energy density of the robot to the welding pool in the current motion state. Based on the instantaneous input energy, the heat conduction model in the welding heat process is used to solve the heat behavior of the molten pool, and a three-dimensional time-space temperature field including thermal physical properties such as thermal diffusion coefficient, metal density and specific heat capacity is constructed. The spatial diffusion behavior and time evolution process of the internal temperature of the molten pool in the welding process are established, and the temperature distribution at different positions and times in the welding area is calculated by numerical solution method to reflect the conduction and accumulation effect of heat between the base material and the molten pool, and the welding pool temperature field is obtained. Based on the welding pool temperature field, the modeling of the surface heat loss mechanism in the welding process is introduced, and the convection and radiation two heat transfer paths are considered. By setting the convective heat transfer coefficient and the metal surface emissivity, combined with the temperature difference between the welding area and the surrounding environment, the total heat loss power of the molten pool outside per unit time is estimated. The convection heat transfer part mainly involves the heat exchange between the molten pool surface and the gas environment, while the radiation part reflects the electromagnetic form of heat energy released by the high-temperature metal surface to the outside space. The instantaneous input energy and the heat loss power are substituted into the energy balance equation to establish the mathematical relationship between the energy supply and demand. In the coupling optimization process, the relationship between different parameters is iteratively adjusted and converged by numerical algorithm, and the optimal control quantity that can meet the heat input requirement of the molten pool and does not produce overheating or underheating is solved, and the balance control parameters obtained include the target heat input level, the current adjustment value, the welding speed adjustment coefficient, etc.

[0023] 104. Calculate joint compensation and power adjustment factor according to balance control parameters and generate first welding control instruction.

[0024] Specifically, during the actual welding process, the running parameters of the tower foot robot in the welding state are continuously collected, and the real-time running data are compared with the balance control parameters calculated in advance through the energy balance optimization module to construct the deviation between the current system running state and the theoretical optimal state. The visual positioning deviation data are obtained by comparing the actual welding point coordinates obtained by the weld position recognition and tracking system with the planned welding path coordinates; at the same time, the speed deviation data are obtained by comparing the current joint radial speed of the robot read by the sensor with the speed corresponding to the optimal motion trajectory; the energy input value composed of the actual welding current and voltage parameters is compared with the balanced energy reference value to obtain the energy deviation data. The three types of deviation data are input into the fuzzy neural network controller, the fuzzy neural network controller performs fuzzy processing on the input signals, maps the continuous numerical error data to multiple language class memberships such as "small deviation", "moderate deviation" and "large deviation" using the Gaussian membership function, and constructs a fuzzy input space through fuzzy rules. In the middle hidden layer of the neural network, the fuzzy input signals are nonlinearly converted by the hyperbolic tangent activation function to form hidden layer neuron outputs with response characteristics to different deviation combinations. The controller has completed the learning of a large number of welding process samples in the training stage, and its internal weight matrix can effectively express the mapping relationship between the visual deviation, speed deviation and energy deviation and the control compensation. After matrix weighting calculation and activation operation, the network outputs two sets of control quantities, in which the joint angle compensation quantity is used to correct the current two-dimensional joint angle mapping angle, and the power adjustment factor is used to adjust the actual welding current intensity according to the energy state. The joint angle compensation quantity is directly superimposed with the two-dimensional joint angle execution value to generate the corrected target joint angle instruction, and the power adjustment factor is applied as a multiplication factor to the current welding current parameter to adjust the output power, thereby jointly constituting the first welding control instruction.

[0025] In a specific embodiment, the process of step 101 can specifically include the following steps: RGB image data of the tower foot angle steel is collected by a binocular camera, and three-dimensional point cloud data of the tower foot angle steel is obtained by laser radar scanning; The three-dimensional point cloud data is subjected to voxel filtering and plane fitting to extract the edge line features of the angle steel; The RGB image data is input into a deep learning network for convolution calculation and full connection layer inference to output a welding bevel type probability distribution; Based on the edge line features of the angle steel and the welding bevel type probability distribution, spatial position fusion is performed to obtain three-dimensional coordinates of the welding point.

[0026] Specifically, the initialization configuration and time synchronization of the sensor system are completed before the start of the tower foot welding operation, so that the binocular vision system and the laser radar system perform data acquisition with a unified reference clock, and a common coordinate system is established in space through a calibration matrix. The binocular camera is arranged on the end of the robot or an independent rack, a high-resolution RGB image acquisition module is used to continuously acquire images of the surface of the tower foot angle steel, the left and right eyes of the camera synchronously shoot image pairs, and the laser radar performs multi-point ranging operation on the same angle steel area in a scanning form to obtain high-density three-dimensional point cloud data. Under a preset sampling angle and scanning frequency, the three-dimensional point cloud data can cover the angle steel profile, the weld area and the surrounding structure. The original point cloud data obtained by the laser radar is preprocessed, a voxel filtering algorithm is used for down-sampling processing of the point cloud, a three-dimensional voxel grid with a fixed size is set, the points falling into the same voxel are replaced by a central point, so that the amount of point cloud data is reduced and the processing efficiency is improved without obvious loss of structure accuracy. The down-sampled point cloud data is geometrically reconstructed by a plane fitting algorithm based on a random sample consensus model, the continuously distributed plane regions in the point cloud are extracted as the surface planes of the angle steel, and a residual threshold and a maximum iteration number are set in the fitting process to filter out the real plane regions and ignore the stray points. The intersection line between two adjacent fitted planes is calculated and the boundary is constructed, and the edge lines reflecting the geometric profile of the angle steel are extracted. These edge lines are the main reference base lines of the weld, and have high welding path constraint value. In the image processing path, the RGB image is input into the trained deep learning network model, the structure of the deep learning network model includes a plurality of convolution layers, pooling layers and full connection layers, the input image extracts low-level features such as color, texture and edge through convolution operation, and then constructs medium and high-level semantic information through multi-layer feature fusion, and the convolution output is mapped to a fixed-dimensional classification vector space through the full connection layer. The classification vector outputs the occurrence probability of each type of welding groove in the current image, forming a probability distribution of the welding types including butt joints, T joints and corner joints. The deep learning network model is trained offline based on a large-scale welding image dataset and deployed in the field edge device. The processing results of the image branch and the point cloud branch are fused in the unified spatial coordinate system, the geometric information of the angle steel edge line and the semantic information of the welding groove type probability distribution in the image are used for position matching and type checking, the two-dimensional pixel position of the welding point in the image and the structure projection relationship of the laser point cloud in the three-dimensional space are combined, the welding groove structure identified in the image is mapped to the three-dimensional space through the fused calibration matrix of the binocular and laser after calibration, the coordinates of the welding groove starting point, midpoint or endpoint are locked, and the welding direction and inclination are judged according to the edge line direction of the angle steel and the groove type. The corresponding relationship between the semantics and the geometry is constructed between the image and the point cloud, and the three-dimensional coordinates of the welding points meeting the welding path planning requirements are output.

[0027] In a specific embodiment, the process of performing step 102 can specifically include the following steps: The motion of the tower foot robot is analyzed in combination with the three-dimensional coordinates of the welding points, the motion planes of the base rotation and the main arm bending are determined, and the motion planes of the base rotation and the main arm bending are taken as two-dimensional mapping dimension reduction parameters; Based on the two-dimensional mapping dimension reduction parameters, the inverse tangent function is calculated to obtain the base rotation angle, and the cosine law is used to calculate the main arm joint angle, and the base rotation angle and the main arm joint angle are taken as the initial joint two-dimensional mapping values of the tower foot robot; The initial joint two-dimensional mapping values are subjected to Jacobian matrix operation to obtain the target joint angle; The visual positioning deviation vector is calculated according to the target joint angle and the welding point target position, and the robot joint two-dimensional angle is generated through the PID controller; Based on the robot joint two-dimensional angle, the radial velocity is calculated to obtain the joint radial velocity.

[0028] Specifically, based on the three-dimensional coordinates of the welding points in space, combined with the task characteristics of the tower foot robot body structure and welding operation, an adaptive motion simplification model is established. In a multi-degree-of-freedom redundant robot, joint calculation involves complex kinematics solving in six-dimensional space. However, the motion in the tower foot welding scene is mainly concentrated in the two-dimensional plane expansion operation, so the projection behavior of the welding point pose in the base plane and the arm expansion direction is extracted, thereby reducing the motion control problem to a two-dimensional parameter space with base rotation and plane expansion as the core. By projecting the three-dimensional coordinates of the welding point on the XY plane, the direction of the robot body from the center point to the welding point is analyzed, and the main motion plane in which the robot base rotates is determined. At the same time, combined with the Z-axis coordinate and the radial horizontal distance, the relative bending degree of the main arm in the three-dimensional structure is derived, and the motion plane in which the main arm bends is determined. These two planes constitute the two-dimensional key action space of the robot. Using the coordinate information of the welding point on the XY plane, the arctangent function is calculated, and the horizontal projection direction from the origin to the target point is used as a reference to calculate the angle at which the robot base needs to rotate, reflecting the deflection range of the welding point in the horizontal position angle direction. The spatial straight-line distance from the welding point to the origin of the robot body is taken as the input, combined with the geometric size of the large arm and the small arm, and the cosine theorem is used to calculate the bending angle of the main arm according to the triangular geometric relationship, describing the specific stretching degree required for the end to reach the welding point under the constraint of the known length member, constituting the first set of two-dimensional joint angle values of the robot for driving the actuator, i.e. the initial two-dimensional joint mapping solution. In order to realize the adjustment of the end to reach the target welding point, the two-dimensional joint angle mapping value is input into the Jacobian matrix operation module, the sensitivity mapping of the joint angle change to the end position change under the current configuration is established, the linear transformation relationship between the robot joint and the end is solved, and the actual target joint angle required for the correction target state is obtained. Based on this, the spatial deviation between the current state and the target state is analyzed. The spatial deviation is obtained by subtracting the actual measurement value of the welding point from the end position derived by the positive solution of the Jacobian matrix, and a three-dimensional positioning deviation vector is constructed, reflecting the positional error of the robot in the current state that fails to accurately reach the welding target. The deviation vector is input into the PID controller for closed-loop compensation control. The PID controller outputs an incremental compensation value for correcting the current two-dimensional joint angle by proportional adjustment, integral accumulation, and differential prediction. The incremental compensation value is added to the initial two-dimensional angle to form a new corrected angle command and is updated to the controller to guide the real-time adjustment of the motion state of the execution system. The updated two-dimensional joint angle is taken as the input variable, combined with the current robot mechanism configuration parameters, the angular velocity of the corresponding joint is calculated, and the overall radial linear velocity is obtained according to the structure projection relationship, which is the actual joint radial velocity during the welding operation execution process.

[0029] In a specific embodiment, the process of calculating the visual positioning deviation vector according to the target joint angle and the target position of the welding point and generating the two-dimensional angle of the robot joint by the PID controller can specifically include the following steps: calculating the actual position coordinates of the robot end effector of the tower foot based on the target joint angle through the forward kinematics equation; performing three-dimensional space difference operation according to the actual position coordinates and the target position of the welding point to obtain the visual positioning deviation vector; inputting the visual positioning deviation vector into the PID controller to perform proportional, integral and differential operation and calculating the joint angle compensation, to obtain the joint compensation control quantity; superimposing and correcting the target joint angle based on the joint compensation control quantity to obtain the two-dimensional angle of the robot joint.

[0030] Specifically, a robot forward kinematics model is constructed with the target joint angle as the input variable. The robot forward kinematics model maps the input joint angle to the absolute position coordinates of the end effector in three-dimensional space based on the standard link coordinate transformation method. By sequentially multiplying the coordinate transformation matrices corresponding to each joint in the robot structure, the original two-dimensional joint angle value, i.e., the base rotation angle and the main arm bending angle, is combined through a set of spatial rotation and translation operations to solve the three-dimensional position of the end effector in space, including the horizontal position, the vertical height and the azimuth angle information of the end effector relative to the base coordinate system, to obtain the actual execution position of the robot end effector under the current input condition. The end position coordinates are compared and analyzed with the target space position of the welding point identified by the vision system in advance, and the three-dimensional visual positioning deviation vector representing the position error between the actual execution point and the theoretical welding point is formed by calculating the difference between the corresponding elements of the three-dimensional coordinates, reflecting the static accuracy deviation of the robot in the position control process, and embodying the cumulative error results of various interference factors such as multi-sensor calibration error, mechanical arm stiffness fluctuation and executor response lag. The three-dimensional visual positioning deviation vector is input into the PID controller for operation. The PID controller directly responds to the deviation size through the proportional term, corrects the deviation accumulation trend through the integral term, and predicts the control of the deviation change rate through the differential term, effectively suppressing the influence of high-frequency noise and improving the stability of the control system. The output of the PID controller is a set of joint angle correction quantities corresponding to the deviation direction, and the selection of the proportional coefficient, integral coefficient and differential coefficient is based on the system identification and controller setting results. The controller outputs a new set of joint angle compensation quantities, i.e., the correction increment of the current target joint angle, according to the real-time deviation in each control period. The joint compensation control quantity is superimposed and operated with the original target joint angle to obtain the real-time updated two-dimensional angle of the robot joint.

[0031] In a specific embodiment, the process of obtaining joint radial velocity based on robot joint two-dimensional angle and radial velocity calculation can specifically include the following steps: Differential calculation is performed on the robot joint two-dimensional angle to obtain joint angular velocity; The joint angular velocity is multiplied by the velocity Jacobian matrix to obtain the radial motion velocity of the welding torch; Based on the radial motion velocity of the welding torch, a radial velocity optimization objective function is established, which includes a joint rotational inertia term, an angular acceleration smoothing term and a velocity tracking term; The radial velocity optimization objective function is solved by using a sequential quadratic programming algorithm to obtain the joint radial velocity.

[0032] Specifically, based on the real-time updated two-dimensional angle sequence of the robot joints, the numerical differentiation method is used to calculate the angle change rate in the continuous time sequence, and the ratio of the angle difference between adjacent time points to the time interval is used to approximate the differential result, respectively obtaining the instantaneous angular velocity of the base rotation joint and the main arm bending joint. The joint angular velocity vector is input into the speed mapping module, and matrix multiplication operation is performed with the velocity Jacobian matrix under the current configuration. The velocity Jacobian matrix is used to express the influence relationship of the joint angular velocity on the linear velocity of the end effector, and its form is developed according to the partial derivative of the end position to each joint angle in the forward kinematics model. Through matrix multiplication, the angular velocity is mapped to the radial motion speed of the welding torch in the Cartesian space. The radial velocity of the welding torch describes the actual moving speed of the welding torch along the current welding direction, which directly affects the welding heat input, the formation of the molten pool and the stability of the weld quality. Based on the radial motion speed of the welding torch, a radial velocity optimization objective function is established, which includes joint rotational inertia term, angular acceleration smoothing term and speed tracking term. The joint rotational inertia term is used to measure the energy consumption of each joint in maintaining the current angular velocity state. The greater the angular velocity or the greater the inertia will result in higher energy consumption. The angular acceleration smoothing term is used to punish the sharp change of joint angular velocity, to suppress the sudden acceleration or deceleration phenomenon, and to improve the continuity and stability of the trajectory execution in the welding process. The speed tracking term is used to ensure that the actual moving speed of the welding torch can be close to the target speed range required by the welding task planning, so as to avoid the welding quality fluctuation caused by the large speed deviation. The joint rotational inertia term, the angular acceleration smoothing term and the speed tracking term are combined by weighted linear combination to form a composite optimization function with multiple objective constraints. In the weight design, the requirements of different welding tasks for energy efficiency, smoothness or accuracy are adjusted. The sequence quadratic programming algorithm is used to optimize the radial velocity optimization objective function. In each iteration, a quadratic approximation model of the objective function near the current solution is constructed, and the gradient information of the current variable and the Lagrange multiplier are used to construct a quadratic programming sub-problem containing the constraint conditions. In each iteration, the curvature information of the objective function is described by the Hessian matrix to determine the search direction and step update method. The new variable update amount is obtained by solving the quadratic sub-problem, and the updated solution is fed back to the original problem for the next round of approximation until the preset convergence criterion is met. The construction of the Hessian matrix includes the second-order derivative information of the objective function and the coupling effect of the constraint term on the variable, so the modified quasi-Newton method is used to approximate and update the Hessian matrix in practical application. After the iteration optimization process of the sequence quadratic programming algorithm, a set of joint angular velocity values are obtained, and the joint radial velocity output that meets the requirements of the welding working condition is obtained.

[0033] In a specific embodiment, the process of performing step 103 can specifically include the following steps: Based on the joint radial velocity, the arc efficiency and the welding voltage and current parameters are divided to obtain the instantaneous input energy; The transient input energy is substituted into the partial differential equation of the molten pool heat conduction containing the thermal diffusion coefficient and the thermal physical parameters of the steel material to obtain the distribution of the temperature field of the welding molten pool through numerical solution; According to the heat transfer calculation of the distribution of the temperature field of the molten pool and the convective heat transfer coefficient and the radiation emissivity, the heat dissipation loss power is obtained; The transient input energy and the heat dissipation loss power are substituted into the energy balance equation to obtain the balance control parameter through coupling optimization solution.

[0034] Specifically, the joint radial velocity of the robot in the current execution state is taken as a basis variable, and real-time current, voltage, and arc efficiency and other process parameters in the welding process are synchronously collected. The actual heat energy injected into the molten pool per unit path length is constructed by multiplying the arc output power by the efficiency factor and dividing by the instantaneous radial velocity of the welding torch, i.e., the welding instantaneous input energy, which has a physical meaning of the energy density corresponding to the current speed of the welding torch per unit time. The heat energy input is substituted into the constructed partial differential model of molten pool heat conduction. The molten pool heat conduction partial differential model takes the weld area as a solution domain, and introduces the thermal physical properties of the steel material such as thermal diffusion coefficient, density, and specific heat capacity. The time is taken as an evolution dimension, and the space is taken as a diffusion dimension to describe the whole process of diffusion and accumulation of the welding input heat in the base material over time. In actual solving, the partial differential equation is discretized by using numerical methods such as finite difference or finite element, the continuous space is divided into fine calculation grids, the numerical solution of the temperature evolution in each unit over time is calculated by combining the heat source injection position of the welding path at each time, and the temperature field distribution of the whole welding area during the welding process is obtained. According to the temperature field distribution of the molten pool, the heat exchange mechanism between the environment and the workpiece is introduced to establish a heat power loss model under the heat transfer path, which includes a forced convection heat loss term between the molten pool surface and the surrounding environment and a high-temperature surface radiation loss term to the surrounding space. The convection heat transfer coefficient is set, which depends on the environmental wind speed, the type and flow rate of the protective gas and other external conditions, and the radiation emissivity of the steel surface is introduced to consider the thermal radiation ability at different temperatures. The heat power released from the molten pool surface to the external environment per unit time is calculated according to the physical law, and the two parts are added to constitute the heat loss term in the welding process. The welding instantaneous input energy and the heat loss power are substituted into the unified energy balance equation for solving to construct the dynamic coupling relationship among energy inflow, outflow, and retention, and the energy balance model is taken as the core to construct the optimization problem. The actual input energy adjustment amount, the current adjustment coefficient, the welding speed correction factor, and the like are taken as optimization variables, the energy conservation constraint and the thermal stability condition are taken as boundary limits, the variable space is searched by the iterative optimization method, and the optimal control combination that can make the heat input, heat diffusion, and heat loss reach dynamic balance under the current speed, current, voltage, and the like is solved, i.e., the balance control parameter, which contains the parameter index that can adjust the actual heat input, is used to feedback drive the power regulation module and the trajectory execution module to realize adaptive regulation and control.

[0035] In a specific embodiment, the process of substituting the instantaneous input energy and the heat loss power into the energy balance equation for coupled optimization solving to obtain the balance control parameter can specifically include the following steps: Subtracting the instantaneous input energy and the heat loss power and establishing an energy balance equation with the product of the molten pool mass and the specific heat capacity; The welding current range constraint and the radial velocity boundary constraint are set based on an energy balance equation; The welding current range constraint and the radial velocity boundary constraint are solved with the welding current and the radial velocity by gradient descent to obtain the balance control parameters.

[0036] Specifically, the instantaneous input energy and the heat loss power are operated by the cycle-by-cycle difference operation, and the difference represents the effective heat energy injected into the molten pool per unit time, that is, the heat source that is really used to melt the base material and maintain the stability of the molten pool in the welding process. In the ideal state, the instantaneous input energy is greater than or equal to the heat loss power, so as to ensure that the welding pool exists continuously and has sufficient penetration, and if the difference is negative, it indicates that the molten pool is in a heat loss state and is insufficient to maintain an effective welding process. The heat energy difference is coupled with the molten pool heat demand model to establish an energy balance equation. In the energy balance equation, the molten pool is regarded as a heat absorber with uniform heat capacity characteristics, and the product of the molten pool mass and the specific heat capacity of the material is introduced as the heat capacity parameter. The product of the heat capacity and the temperature rise is the energy accumulation required by the molten pool. By equating the net energy obtained by subtracting the heat loss power from the input heat and the molten pool heat capacity term, a mathematical model reflecting the balance relationship between the temperature change rate of the molten pool and the dynamic of heat input is established, which reflects the dynamic transformation process between the injection, conduction, consumption and accumulation of heat energy in the welding process. To avoid the problem of energy imbalance leading to the temperature of the molten pool fluctuating sharply or the penetration being insufficient, the control variables involved in the energy balance equation are set with constraints. The welding current, as a direct adjustment parameter of energy input, is limited between the minimum and maximum values allowed by the process, and the range depends on the thickness of the welding material, the welding process specification and the requirements of the weld structure, while the radial speed reflects the moving speed of the robot welding path and directly affects the heat density absorbed per unit length of the weld, so the boundary conditions are set to prevent the weld from not being fused due to too fast or overheating and burning through due to too slow. The above constraints together form the boundary conditions of the nonlinear optimization problem, and the control system will find the optimal variable combination within the boundary. To solve the nonlinear optimization problem, the gradient descent algorithm is introduced as the main optimization method. By taking the current and radial speed as the variables to be optimized, the energy balance deviation function is defined as the loss function, and the gradient direction is constructed by taking the partial derivative of the function with respect to the current and speed. In each iteration, the instantaneous input energy, heat loss power and their difference are calculated according to the current and speed combination, and the molten pool heat state deviation is evaluated by substituting them into the energy balance equation. According to the deviation direction, the current and speed are adjusted to make the system heat input tend to the molten pool heat demand, so as to drive the energy balance equation to approach zero deviation direction. In the gradient descent process, to avoid jumping out of the boundary, the projected gradient method or the gradient descent strategy with constraints is used to limit the current value within the preset current range and the speed value within the allowed radial speed boundary interval after each variable update, so as to ensure that the optimization variables always meet the process constraint conditions. When the gradient descent algorithm converges to a stable solution within an acceptable error range after several iterations, the current and radial speed combination is determined as the balanced control parameter set that meets the energy balance, heat dissipation control and molten pool stability, wherein the current control parameter is used to adjust the output power of the welding power supply, and the speed control parameter is used to correct the execution rate of the robot path.

[0037] In a specific embodiment, the process of performing step 104 can specifically include the following steps: Based on the balance control parameters and the real-time motion state of the tower foot robot, difference operation is performed to obtain visual positioning deviation data, speed deviation data and energy deviation data; The visual positioning deviation data, speed deviation data and energy deviation data are input into the fuzzy neural network for Gaussian membership function fuzzification processing to obtain three fuzzy input signals; Based on the three fuzzy input signals, hyperbolic tangent activation operation and weight matrix calculation are performed to obtain joint compensation and power adjustment factor; The joint compensation is superimposed with the two-dimensional angle of the robot joint, and the power adjustment factor is adjusted with the welding current to obtain the first welding control instruction.

[0038] Specifically, a dynamic difference relationship between the balance control parameters and the current actual motion state of the robot is established, wherein the balance control parameters are a set of target values calculated offline or online by energy optimization and speed planning models, covering the end position coordinates, joint radial velocity and input heat energy level expected to be maintained during the welding process, and the real-time motion state is derived from the actual values collected by sensors, encoders, vision systems and power detection devices during the operation of the robot. The target end position and the actual welding point coordinates are subjected to spatial coordinate difference operation to obtain visual positioning deviation data, reflecting the instantaneous error of the robot in terms of spatial tracking accuracy; the target radial velocity and the current joint angular velocity are subjected to difference calculation to obtain speed deviation data, which represents the consistency degree of trajectory execution dynamics; at the same time, the actual power is calculated by input energy and heat feedback, and the energy injection amount per unit time is calculated by difference operation with the expected energy input level to obtain energy deviation data, which is used to evaluate the stability of the heat control closed loop. The visual positioning deviation data, speed deviation data and energy deviation data are input into the fuzzy neural network for fuzzy processing, and the Gaussian membership function is introduced as a fuzzy operator. The Gaussian membership function takes the center value and the standard deviation as parameters, converts the continuous real error value into a membership distribution with fuzzy characteristics, and sets several fuzzy sets such as "low", "medium", "high" and "too high" for each input channel, and maps the error data to the membership values corresponding to each level through the function to form three fuzzy input signal channels. The fuzzy input signals enter the hidden layer processing stage of the fuzzy neural network, are input into the neural calculation unit with nonlinear activation, the hyperbolic tangent function is used as the activation function, the input signals have continuous differentiable response ability in the positive and negative range, and the joint state of different input signals is weighted and combined through the preset weight matrix in the network structure to form a group of hidden layer neuron outputs with nonlinear mapping characteristics, reflecting the response mode of the system under the current deviation combination, and generating two groups of output data through neural connection mapping, wherein the joint compensation amount is used to correct the current two-dimensional joint angle to control the path tracking error, and the power adjustment factor is used to dynamically adjust the welding current to maintain the stability of the heat input. The joint compensation amount is directly superimposed with the current calculated two-dimensional joint angle value of the robot to construct a new angle control instruction for driving the robot executor to adjust the posture to correct the end position; at the same time, the power adjustment factor is used as a proportional coefficient to act on the current current setting value, and an adjusted current control value is output to ensure that the welding current is automatically increased or decreased when there is a heat input deviation, and dynamic heat compensation is realized. The angle control instruction and the current control value are combined to generate a first welding control instruction, which includes two parts of robot motion instruction and welding power instruction, acting on the motion control layer and the welding control layer respectively.

[0039] In a specific embodiment, the method for adaptive visual cooperative tower foot robot welding further comprises the following steps: Projecting a line structured light on the weld surface during the tower foot welding process and collecting images to obtain weld contour deformation data; Calculating the weld cross-sectional geometric dimensions based on the weld contour deformation data and extracting weld geometric feature data; Quality assessment of the weld geometric feature data to obtain quality evaluation parameters; Adjusting the joint compensation amount and power adjustment factor in the first welding control instruction according to the quality evaluation parameters through the gradient ascent algorithm to generate the second welding control instruction.

[0040] Specifically, a line structured light scanning system is configured in the process of welding tower foot, which includes a laser line projection module and a high-resolution CCD camera. At the end of welding or in the cooling stage after the weld is formed, a high-brightness line laser light strip is projected onto the weld surface. Due to the height fluctuation of the weld geometry, the laser line will produce regular deformation on the weld surface, which contains key weld structure information such as height, width, excess height, and depression. At the same time, the CCD camera collects images of the laser line from a fixed angle and transmits them to the processing system in real time. Based on the triangular geometric relationship between the laser line offset and the camera view angle, the image coordinates are converted to actual spatial coordinates using the spatial back-projection technique to construct the cross-sectional height data profile of the weld surface. Based on the obtained weld profile deformation data, an analysis model for cross-sectional geometric feature extraction is constructed. The analysis model projects the collected profile data onto the vertical plane of the weld cross-section and performs edge detection and geometric fitting operations to calculate the typical size parameters of the weld cross-section, including weld width, weld excess height, undercut depth, and fusion angle. In the specific calculation process, the weld width is determined by identifying the boundary points of the profile and calculating the horizontal distance between the two sides of the base material. The excess height is calculated by analyzing the vertical height of the highest point in the profile relative to the reference plane. If a concave pit region is detected on the profile edge, the undercut depth is measured. At the same time, the geometric continuity at the weld toe is analyzed to judge the fusion quality. The weld geometric feature data is input into the quality evaluation model for comprehensive analysis. The quality evaluation model establishes an evaluation function based on the standard geometric specifications of the weld, compares the deviation between the actual geometric size of the weld and the target weld standard value, and calculates the weld quality evaluation parameter. The weld quality evaluation parameter has a value range of 0 to 1, where a value close to 1 indicates excellent weld quality, and a value close to 0 represents the presence of serious welding defects. Exponential or weighted loss functions are introduced into the quality evaluation model, and weight factors are set for typical problems such as weld width deviation, excess height deficiency, and undercut depth. The exponential function suppresses the influence of small errors and amplifies the influence of serious defects, improving the sensitivity of the evaluation result to key defects and outputting a unified quality score. The quality evaluation result is used as feedback input to guide the self-optimization of the welding control parameters. Based on the gradient ascent algorithm, the current control parameters are corrected. Due to the nonlinear response relationship between the weld quality and the control command, the gradient ascent algorithm adjusts the control quantity in the direction of the fastest quality function value increase. In the calculation process, the gradient vectors of the quality function are constructed for joint compensation and power adjustment factors. The change rate of the quality score under a small perturbation of the current control parameters is calculated by numerical differentiation method, and the control quantity in the original first welding control command is incrementally adjusted according to the gradient direction. A new set of compensation and adjustment factors is updated in each control cycle. Through continuous iteration, when the weld quality score reaches or approaches the set quality threshold, it is determined that the control command is effective and stable. If the quality score still does not meet the standard, the gradient correction continues.The second welding control instruction is composed of the updated compensation amount and the adjustment factor.

[0041] The adaptive visual collaborative tower leg robot welding method in the embodiment of the application is described above, and the adaptive visual collaborative tower leg robot welding system in the embodiment of the application is described below, please refer to Figure 2 The adaptive visual collaborative tower leg robot welding system in the embodiment of the application includes one embodiment: The composite scanning module 201 is configured to perform binocular vision and laser radar composite scanning on the tower leg angle steel to obtain three-dimensional coordinates of the welding points; The coordinate mapping conversion module 202 is configured to perform joint two-dimensional coordinate mapping conversion on the three-dimensional coordinates of the welding points to obtain robot joint two-dimensional angles and calculate joint radial velocities; The welding energy balance calculation module 203 is configured to perform welding energy balance calculation according to the joint radial velocities to obtain balance control parameters; The generation module 204 is configured to calculate joint compensation amounts and power adjustment factors according to the balance control parameters and generate first welding control instructions.

[0042] Through the cooperation of the above-mentioned components, the RGB image information obtained by the binocular camera is combined with the high-precision three-dimensional point cloud data provided by the laser radar to form a visual-geometric dual-constraint recognition mechanism, effectively overcoming the problem that a single visual system is easily disturbed by welding splashes and smoke, and improving the recognition accuracy and anti-interference ability of the tower leg angle steel welding points. The traditional six-dimensional joint space inverse kinematics solving is reduced to a two-dimensional key motion plane of base rotation and main arm bending, which greatly reduces the calculation complexity, eliminates the multi-solution and no-solution problems, establishes a coupling optimization model of robot joint radial velocity and welding energy input, and synchronously optimizes the motion parameters and process parameters through the sequential quadratic programming algorithm, which solves the low energy utilization problem caused by independent optimization of motion control and welding process in the traditional method. The fuzzy neural network fuses visual feedback, motion state and energy state information, dynamically adjusts the control parameters through an online learning mechanism, so that the tower leg robot welding system can adapt to different thickness angle steels and changing assembly gaps, and the welding quality stability is improved. The laser contour scanning is used to monitor the weld geometry in real time, and the gradient ascent algorithm is used for parameter feedback adjustment, so as to ensure the quality consistency and controllability of the tower leg welding process.

[0043] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described system, system and unit can refer to the corresponding processes in the foregoing method embodiments, which will not be described herein.

[0044] The integrated unit, if implemented in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application or the entire or part of the technical solutions that essentially contribute to the prior art can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0045] The above-described embodiments are merely used to illustrate the technical solutions of the present application, rather than limit the same; even though the present application has been described in detail with reference to the foregoing embodiments, those ordinarily skilled in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements for some of the technical features; and these modifications or replacements do not cause the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. An adaptive vision-coordinated tower leg robot welding method, characterized in that: include: Perform binocular vision and laser radar composite scanning on the tower foot angle steel to obtain the three-dimensional coordinates of the welding points; The three-dimensional coordinates of the welding point are converted into two-dimensional coordinates of the joint to obtain the two-dimensional angle of the robot joint and calculate the radial velocity of the joint; Perform welding energy balance calculation based on the joint radial velocity to obtain balance control parameters; The joint compensation amount and the power adjustment factor are calculated according to the balance control parameters and a first welding control instruction is generated.

2. The adaptive vision-coordinated tower leg robot welding method according to claim 1 is characterized in that: The method of performing binocular vision and laser radar composite scanning on the tower foot angle steel to obtain the three-dimensional coordinates of the welding points includes: The RGB image data of the tower foot angle steel is collected by a binocular camera, and the three-dimensional point cloud data of the tower foot angle steel is obtained by laser radar scanning; Performing voxel filtering and plane fitting on the three-dimensional point cloud data to extract edge line features of the angle steel; Input the RGB image data into a deep learning network for convolution calculation and full connection layer reasoning, and output a probability distribution of welding groove types; Based on the edge line features of the angle steel and the probability distribution of the welding groove type, spatial position fusion is performed to obtain the three-dimensional coordinates of the welding point.

3. The adaptive vision-coordinated tower leg robot welding method according to claim 1 is characterized in that: The three-dimensional coordinates of the welding points are converted into two-dimensional coordinates of the joints to obtain two-dimensional angles of the robot joints, and the radial velocity of the joints is calculated, including: Performing motion analysis on the tower foot robot based on the three-dimensional coordinates of the welding points to determine the motion planes of the base rotation and the main arm bending, and using the motion planes of the base rotation and the main arm bending as two-dimensional mapping dimensionality reduction parameters; An inverse tangent function is performed based on the two-dimensional mapping dimension reduction parameter to obtain a base rotation angle, and a main arm joint angle is calculated using the cosine theorem, and the base rotation angle and the main arm joint angle are used as initial joint two-dimensional mapping values ​​of the tower foot robot; Performing a Jacobian matrix operation on the initial joint two-dimensional mapping value to obtain a target joint angle; Calculating a visual positioning deviation vector based on the target joint angle and the target position of the welding point and generating a two-dimensional angle of the robot joint through a PID controller; The radial velocity of the robot joint is calculated based on the two-dimensional angle of the robot joint to obtain the radial velocity of the joint.

4. The adaptive vision-coordinated tower leg robot welding method according to claim 3 is characterized in that: The method of calculating the visual positioning deviation vector according to the target joint angle and the target position of the welding point and generating the robot joint two-dimensional angle through the PID controller includes: Calculate the actual position coordinates of the end effector of the tower foot robot through the forward kinematics equation based on the target joint angle; Performing a three-dimensional spatial difference operation on the actual position coordinates and the target position of the welding point to obtain a visual positioning deviation vector; Inputting the visual positioning deviation vector into the PID controller to perform proportional integral differential operation and calculate the joint angle compensation amount to obtain the joint compensation control amount; The target joint angle is corrected based on the joint compensation control amount to obtain a two-dimensional angle of the robot joint.

5. The adaptive vision-coordinated tower leg robot welding method according to claim 4 is characterized in that: The radial velocity calculation based on the two-dimensional angle of the robot joint to obtain the joint radial velocity includes: Performing differential calculation on the two-dimensional angles of the robot joints to obtain the angular velocity of each joint; Performing a product operation on the angular velocity of each joint and the velocity Jacobian matrix to obtain the radial motion velocity of the welding gun; Establishing a radial velocity optimization objective function including a joint moment of inertia term, an angular acceleration smoothing term, and a velocity tracking term based on the radial motion velocity of the welding gun; The radial velocity optimization objective function is solved by Hessian matrix iteration using a sequential quadratic programming algorithm to obtain the joint radial velocity.

6. The adaptive vision-coordinated tower leg robot welding method according to claim 5 is characterized in that: The welding energy balance calculation is performed according to the joint radial velocity to obtain the balance control parameters, including: Performing a division operation based on the joint radial velocity, arc efficiency, and welding voltage and current parameters to obtain instantaneous input energy; Substituting the instantaneous input energy into the partial differential equation of heat conduction in the molten pool including the thermal diffusivity and the thermophysical properties of the steel material for numerical solution, the temperature field distribution of the welding molten pool is obtained; Perform heat transfer calculation based on the molten pool temperature field distribution, convection heat transfer coefficient, and radiation emissivity to obtain heat dissipation loss power; The instantaneous input energy and the heat dissipation loss power are substituted into the energy balance equation for coupled optimization solution to obtain the balance control parameters.

7. The adaptive vision-coordinated tower leg robot welding method according to claim 6 is characterized in that: Substituting the instantaneous input energy and the heat dissipation loss power into the energy balance equation for coupled optimization solution to obtain the balance control parameters includes: Subtracting the instantaneous input energy from the heat dissipation loss power and establishing an energy balance equation with the product of the molten pool mass and specific heat capacity; Setting welding current range constraints and radial velocity boundary constraints based on the energy balance equation; The welding current range constraint and the radial velocity boundary constraint are solved by gradient descent with the welding current and radial velocity to obtain balance control parameters.

8. The adaptive vision-coordinated tower leg robot welding method according to claim 1 is characterized in that: The step of calculating the joint compensation amount and the power adjustment factor according to the balance control parameter and generating the first welding control instruction includes: Performing difference calculation based on the balance control parameters and the real-time motion state of the tower foot robot to obtain visual positioning deviation data, speed deviation data and energy deviation data; Inputting the visual positioning deviation data, the speed deviation data and the energy deviation data into a fuzzy neural network for fuzzification processing using a Gaussian membership function to obtain three fuzzified input signals; Performing a hyperbolic tangent activation operation and a weight matrix calculation based on the three-way fuzzified input signal to obtain a joint compensation amount and a power adjustment factor; The joint compensation amount and the two-dimensional angle of the robot joint are superimposed and calculated, and the power adjustment factor and the welding current are adjusted and calculated to obtain a first welding control instruction.

9. The adaptive vision-coordinated tower leg robot welding method according to claim 1 is characterized in that: The adaptive visual collaborative tower leg robot welding method further includes: Line structured light projection and image acquisition are performed on the weld surface during the tower foot welding process to obtain weld contour deformation data; Calculating weld cross-sectional geometric dimensions and extracting weld geometric feature data based on the weld profile deformation data; Performing quality assessment on the weld geometric feature data to obtain quality evaluation parameters; The joint compensation amount and the power adjustment factor in the first welding control instruction are adjusted according to the quality evaluation parameter through a gradient ascent algorithm to generate a second welding control instruction.

10. An adaptive vision-coordinated tower leg robot welding system, characterized in that: The method for performing the adaptive visual collaborative tower leg robot welding method according to any one of claims 1 to 9 comprises: The composite scanning module is used to perform binocular vision and laser radar composite scanning on the tower foot angle steel to obtain the three-dimensional coordinates of the welding points; A coordinate mapping conversion module is used to convert the three-dimensional coordinates of the welding point into two-dimensional coordinates of the joint, obtain the two-dimensional angle of the robot joint, and calculate the radial velocity of the joint; a welding energy balance calculation module, configured to perform welding energy balance calculation according to the joint radial velocity to obtain balance control parameters; A generation module is used to calculate the joint compensation amount and the power adjustment factor according to the balance control parameters and generate a first welding control instruction.

Images