Search

CN-121981961-A - Intelligent extraction method for three-dimensional reconstruction and weld track of complex steel structural member

CN121981961ACN 121981961 ACN121981961 ACN 121981961ACN-121981961-A

Abstract

The invention discloses a three-dimensional reconstruction and weld track intelligent extraction method for a complex steel structural member, and belongs to the technical field of intelligent welding and machine vision. The method comprises the steps of obtaining original point cloud data of a complex steel structural member, preprocessing to obtain a regular point cloud data set, calculating Gaussian curvature and average curvature of each point under different neighborhood scales by adopting a principal curvature analysis method, constructing a multi-scale curvature feature descriptor, carrying out clustering analysis, carrying out feature matching by combining a pre-trained weld geometric model library, identifying candidate weld areas, carrying out area growth and boundary tracking processing on the candidate weld areas, obtaining a smooth continuous weld contour curve through three-time B spline curve fitting, calculating a welding path point sequence and welding gun posture parameters according to the weld contour curve, and generating welding track data which can be executed by a robot. The method can realize the accurate identification of the multiple types of welding seams, generate a smooth and optimized welding track, and has the characteristics of reasonable track planning and accurate real-time tracking.

Inventors

  • HU XIAODONG

Assignees

  • 武汉汉光钢品建设工程有限公司

Dates

Publication Date
20260505
Application Date
20251229

Claims (10)

  1. 1. A three-dimensional reconstruction and weld track intelligent extraction method for a complex steel structural member is characterized by comprising the following steps: acquiring original point cloud data of a complex steel structural member, and preprocessing the original point cloud data to obtain a regular point cloud data set; Based on the regular point cloud data set, calculating Gaussian curvature and average curvature of each point under different neighborhood scales by adopting a principal curvature analysis method, and constructing a multi-scale curvature feature descriptor; Performing cluster analysis on the multi-scale curvature feature descriptors by using a density-based spatial clustering algorithm, and performing feature matching by combining a pre-trained weld geometric model library to identify candidate weld areas; performing region growing and boundary tracking treatment on the candidate weld joint region, and obtaining a smooth continuous weld joint contour curve through three times of B spline curve fitting; and calculating a welding path point sequence and corresponding welding gun posture parameters according to the welding line profile curve, and generating welding track data executable by a robot.
  2. 2. The method for three-dimensional reconstruction and weld track intelligent extraction of complex steel structural members according to claim 1, wherein the preprocessing of the original point cloud data comprises statistical filtering denoising, and the specific process is as follows: calculating the average distance between each point in the point cloud and K nearest neighbors, counting the average distance distribution of all points, judging the points with the average distance exceeding the global average distance plus or minus n times the standard deviation range as noise points, and eliminating the noise points, wherein the value range of K is 20-50, and the value range of n is 1.0-2.0.
  3. 3. The method for three-dimensionally reconstructing a complex steel structural member and intelligently extracting weld trajectories according to claim 1, wherein the method for constructing a multi-scale curvature feature descriptor is as follows: selecting m different neighborhood radii For each point p in the point cloud, fitting a local curved surface under each neighborhood scale by a principal component analysis method, and calculating a corresponding principal curvature And Calculating Gaussian curvature from principal curvature And average curvature Combining curvature values under each scale to form a multi-scale curvature characteristic vector with 2m dimensions, wherein the value range of m is 3-5.
  4. 4. The method for three-dimensional reconstruction and weld track intelligent extraction of complex steel structural members according to claim 1, wherein a density-based spatial clustering algorithm adopts a self-adaptive DBSCAN algorithm, and the parameter determination method comprises the following steps: and adaptively calculating the neighborhood radius epsilon of each point according to the local density characteristic of the point cloud, and automatically determining the global minimum sample number through the inflection point of the K-distance curve, wherein the local density characteristic is represented by adopting the inverse weighted average value of the distances between the point and K nearest neighbors of the point.
  5. 5. The three-dimensional reconstruction and weld track intelligent extraction method for the complex steel structural member is characterized in that a pre-trained weld geometric model library comprises a V-shaped groove weld model, a fillet weld model, a butt weld model and a T-shaped weld model, and the feature matching adopts an improved iterative nearest point algorithm to evaluate the matching degree by calculating the mahalanobis distance between the candidate region feature vector and the feature vector of each weld model.
  6. 6. The method for three-dimensional reconstruction and weld track intelligent extraction of complex steel structural members according to claim 1, wherein the specific process of region growth is as follows: The method comprises the steps of taking a high curvature point as a seed point, carrying out region expansion according to curvature similarity and a normal vector included angle threshold between adjacent points, and taking the point into a welding line region when the curvature difference between the adjacent points is smaller than a set threshold delta kappa and the normal vector included angle is smaller than a set threshold theta, wherein the delta kappa is in a value range of 0.01-0.05, and the theta is in a value range of 10-30 degrees.
  7. 7. The method for three-dimensionally reconstructing a complex steel structural member and intelligently extracting weld trajectories according to claim 1, wherein the method for calculating the posture parameters of the welding gun is as follows: According to the position and tangential vector of each sampling point on the contour curve of the welding seam, the advance angle alpha and the working angle beta of the welding gun are determined according to the welding process requirement, wherein the advance angle alpha is the included angle of projection of the axis of the welding gun and the welding direction in a welding plane, the value range is 70-90 degrees, and the working angle beta is the included angle of the axis of the welding gun and the normal vector of the surface of a workpiece, and the value range is 35-55 degrees.
  8. 8. The method for three-dimensional reconstruction and weld trajectory intelligent extraction of a complex steel structure according to claim 1, further comprising a welding trajectory optimization step of: And performing global optimization on the generated welding track by adopting a self-adaptive genetic algorithm, and establishing an optimization model for double targets by taking the shortest total length of the welding track and the optimal joint motion smoothness as the optimal, wherein the cross probability and the variation probability of the self-adaptive genetic algorithm are dynamically adjusted according to population fitness distribution.
  9. 9. The method for three-dimensional reconstruction and weld trajectory intelligent extraction of a complex steel structure according to claim 8, wherein the fitness function of the optimization model is: ; Wherein L is the total length of the welding path corresponding to the current individual, S is the time integral value of the square of the joint angular acceleration, And The path length and smoothness values corresponding to the initial trajectory, And Is a weight coefficient, satisfies 。
  10. 10. The method for three-dimensional reconstruction and weld track intelligent extraction of complex steel structural members according to claim 1, further comprising the step of weld real-time tracking: In the welding process, position information of a welding seam is acquired in real time through a laser displacement sensor, left-right deviation and high-low deviation of a welding gun and the center of the welding seam are detected, the pose of the welding gun is adjusted in real time according to the deviation information by adopting a fuzzy PID control algorithm, so that self-adaptive tracking of the welding seam is realized, the input of the fuzzy PID controller is a deviation value e and a deviation change rate e c , and correction amounts delta Kp, delta Ki and delta Kd of PID parameters are output.

Description

Intelligent extraction method for three-dimensional reconstruction and weld track of complex steel structural member Technical Field The invention belongs to the technical field of intelligent welding and machine vision, and particularly relates to a three-dimensional reconstruction and weld track intelligent extraction method for a complex steel structural member. The method comprehensively utilizes a laser three-dimensional scanning technology, a point cloud data processing algorithm, a machine learning identification method and a robot track planning technology to realize automatic identification of the welding seam of the complex steel structural member and intelligent generation of the welding track. Background With the rapid development of modern industrial manufacturing to an intelligent direction, welding robots are increasingly widely applied to the fields of heavy industry such as ship manufacturing, bridge construction, pressure vessel production and the like. The complex steel structural member has the characteristics of complex geometric shape, various weld types, uncertain spatial positions and the like, and higher technical requirements on an automatic welding system are provided. The traditional teaching reproduction type welding method requires an operator to manually teach each welding seam, so that the efficiency is low, and the method is difficult to adapt to the production modes of small batches and multiple varieties. Therefore, how to realize automatic recognition of the welding seam and intelligent planning of the welding track becomes a key bottleneck for restricting the improvement of the automation level of welding. In the prior art, a welding seam identification method based on visual sensing mainly comprises a structured light method, a laser scanning method, a binocular vision method and the like. Structured light methods acquire three-dimensional information by projecting a known pattern onto a workpiece surface and analyzing its distortion, but are prone to interference on highly reflective metal surfaces. The laser scanning method has the advantages of high measurement precision and strong anti-interference capability, but the single-line laser scanning has the problem of limited information quantity. Binocular vision relies on feature point matching, and the matching success rate for metal surfaces with sparse textures is low. In the aspect of point cloud data processing, the existing method mostly adopts curvature characteristics of a single scale to carry out weld joint identification, so that geometric characteristics of the weld joint and a workpiece are difficult to effectively distinguish, and the false identification rate is high. In addition, for multiple types of welds on complex steel structures, there is a lack of a uniform identification framework and adaptive parameter adjustment mechanisms. In the aspect of track planning, the existing method mostly focuses on the generation of path points, ignores the optimization of welding gun gestures and the consideration of robot kinematics constraint, and causes the phenomenon of accessibility problem or unstable movement in actual welding. In view of the above technical problems, there is a need to develop a comprehensive solution that can adapt to the geometric characteristics of complex steel structural members, has intelligent recognition capability of multiple types of welding seams, and can generate optimized welding tracks. Disclosure of Invention Aiming at the defects of the prior art, the invention aims to provide a three-dimensional reconstruction and weld track intelligent extraction method for a complex steel structural member. The method comprises the steps of constructing a multi-scale curvature characteristic descriptor, combining a self-adaptive clustering algorithm and a pre-training weld model library, accurately identifying various welds on a complex steel structural member, obtaining a smooth weld contour through three-time B-spline curve fitting, automatically generating a robot executable welding track by combining welding process constraints, performing global optimization on the track by further adopting an intelligent optimization algorithm, and introducing a real-time tracking mechanism to ensure welding quality. In order to achieve the purpose, the invention adopts the following technical scheme that the three-dimensional reconstruction and weld track intelligent extraction method for the complex steel structural member comprises the following steps: acquiring original point cloud data of a complex steel structural member, and preprocessing the original point cloud data to obtain a regular point cloud data set; Based on the regular point cloud data set, calculating Gaussian curvature and average curvature of each point under different neighborhood scales by adopting a principal curvature analysis method, and constructing a multi-scale curvature feature descriptor; Performing cluster analysis o