Research On The Extract And Classification Method Of Welded Seam Defects Based On Point Cloud Data Processing

In the industrial welding production process,welding defects are easily generated due to manual operations and other reasons.With the development of artificial intelligence,the detection of weld defects using equipment such as depth cameras and CMOS cameras in the field of machine vision has achieved better recognition and classification results,but compared to LIDAR,the former is more sensitive to light sources in the harsh environment of welding operations and easily affects the quality of data acquisition,while the latter has strong anti-interference capability,does not depend on external lighting conditions and can work around the clock,through LIDAR point cloud data can provide rich geometric information and display intuitive,so the LIDAR as a measurement device,and the use of point cloud data processing methods to study the extraction and classification of weld defects has some practical significance.The main research contents of this paper include.(1)weld defect three-dimensional measurement system modeling and analysisBy analyzing the type and characteristics of the welded workpiece,the weld object is divided into two categories: linear and three-dimensional,and two workpiece scanning methods,translation and rotation,are designed,and the measurement system as a whole is modeled.For the external parameter calibration between the lidar and the turntable in the rotary scanning,this paper uses the standard ball as the target,and the turntable drives the target rotation to obtain the coordinate correspondence between the lidar coordinate system and the target in the turntable coordinate system at different positions,and then solves the coordinate conversion parameters.(2)Point cloud data acquisition and pre-processingIn this paper,the data acquisition accuracy of the measurement system is analyzed and the optimal scanning parameters of the lidar are selected before collecting the weld point cloud data.For the weld tumor,porosity and burn-through defects studied in this paper,translational scanning was selected to acquire the data.Due to the large number of noisy points in the original point cloud and the large number of points,noise reduction,downsampling and partial removal of the weld plate were performed using filtering methods.In practice a large number of welding seam samples have certain difficulties,so this paper through the rotation transformation and Gaussian noise method for data expansion of the classification samples.(3)Extraction and classification methods for weld defectsIn order to improve the accuracy of weld defect extraction,this paper proposes a point cloud alignment method based on regularized processing,and the defect parts are extracted from the aligned defect point cloud using the double threshold separation method.SVM is selected as the classifier for the classification of weld defects according to the characteristics of the small number and types of samples of the research object,and the geometric features of the defective parts and the global features of VFH are described.(4)External parameter calibration and weld defect extraction and classification experimentsThe experimental results show that the accuracy of the calibration method proposed in this paper is maintained within 0.1mm,and the overall classification accuracy of weld defects reaches 87.16%,which is an effective method for industrial weld defect detection.