Studies Of Key Technologies For Surface Defect Extraction In Automobile Dies

Feature extraction of point clouds is deeply researched in the field of reverse engineering.It can be widely used in reverse design and development of fast products,automobile body panels and decorative parts,design and repair to key parts of aircraft and other industrial designs.Data segmentation,re-sampling,registration and surface defect determination of point clouds all depend on the results of feature extraction.Therefore,it is meaningful both in science and application to study the key technologies for surface defect extraction in automobile dies.In this paper,theoretical and technical research is carried out to extract the surface defects in automobile dies.In order to ensure the accuracy of point cloud segmentation,feature extraction and data registration,the denoising algorithms of point clouds are firstly studied,then the denoised point clouds are segmented into regions,and the feature extraction methods of point clouds are studied on this basis.Finally,the location recognition and feature extraction of surface defects in automobile dies are realized.The main research contents of this paper are as follows:1.Considering the denoising requirements of different regions of point clouds,a feature information classification approach based on the normal distance of sampling points is proposed by analyzing the regional feature information of point clouds.It is a hybrid denoising approach based on feature classification,in which different denoising measures are respectively taken to deal with the smooth regions with fewer features and the sharp regions with more features.Some point cloud models with different intensities of noise are selected to evaluate the effectiveness of the proposed approach.The results show that the approach can enhance the smoothness of the smooth regions,maintain the geometric characteristics of the sharp regions,and avoid excessive smoothness and detail distortion.In terms of precision and computational efficiency,the proposed approach shows its superiority.The advantages of the proposed approach in precision and computational efficiency are verified by comparing the deviation of denoising results and the time of different denoising algorithms.2.Aiming at low efficiency and precision of fuzzy clustering,a hybrid fuzzy region clustering algorithm based on fuzzy clustering and swarm intelligence optimization is proposed.Firstly,social particle swarm optimization(SPSO)algorithm is constructed by introducing the follow characteristics of social atoms based on particle swarm optimization(PSO)algorithm,then an improved social particle swarm optimization(ISPSO)algorithm is designed by using self-adaptive adjustment strategy to optimize the constant follow threshold.Finally,the ISPSO algorithm is used to optimize the fuzzy c-means algorithm to obtain an improved social particle swarm optimization fuzzy c-means(ISPSO-FCM)hybrid fuzzy region clustering algorithm.Some point cloud models with different surface complexity are selected to verify the feasibility of the proposed algorithm.The results show that the hybrid fuzzy region clustering algorithm has good clustering performance.Moreover,it has benefits in clustering accuracy,stability and convergence speed.3.According to the results of region clustering segmentation and the idea of edge detection in image processing,a feature extraction method of point clouds based on region clustering segmentation is introduced by analyzing the neighborhood information and geometric characteristics of point clouds and designing the solution to clustering,extraction,feature point set refinement,segmentation and sorting of point cloud feature information.Point clouds after clustering can maintain the corresponding relationship between the parameter lines of the curved surface and the geometric features of the local area so that the results of feature extraction are more accurate.Feature extraction experiments on several point cloud models with different densities and noise intensities are verified that the proposed algorithm has low sensitivity to noise points,neighborhood scale or sampling quality,and has high accuracy and practicability.4.In order to repair the defects of dies,a surface defect detection method based on registration is proposed,which can be applied to the defect detection of sags and folds in point clouds.Firstly,the two-level registration method based on the four-points congruent algorithm and improved iterative closest point algorithm is used to point clouds registration.After registration,deviation analysis of point clouds is carried out to locate and identify the defect regions.And the hybrid fuzzy clustering algorithm is used to segment the defect regions.Finally,features of defect regions are extracted by using the multi-scale directed line segment angles difference method.And the features are parameterized to identify the types of defect regions.The pillar and insert dies(T059)from Jiangling automobile are applied to the experiments on surface defect detection in automobile dies.The experimental results show that the defects to be repaired are separated and the defect feature parameters are given,which provide the quantitative basis of repairing and processing the dies later.