Research On Complex Free-form Surface Parts Quality Inspection With 3D Registration Method

Free-form surface parts have a wide range of applications in the field of mechanical manufacturing.The design,processing and inspection of such parts involve a lot of sophisticated technologies.Particularly,the aerospace,shipbuilding,automotive and other mechanical processing put forward more stringent requirements on the precision and efficiency of machining parts.In the whole proces of the parts,non-uniform rational B-splines method can solve the design problem perfectly.The high-degree-of-freedom CNC machine tools and rapid prototyping technology can solve the processing problem of such parts with high efficiency.However,as for quality inspection of free-form surface parts,how to evaluate the quality of their manufacturing is still a problem that not be completely resolved.Because of the complexity of the parts structure,there is not a unified technology to complete the effective inspection of free-form surface parts.In recent years,with the rapid development of computer vision and pattern recognition,non-contact measurement technology becomes more and more popular in this field.In particular,the free-form surface optimal registration method is an important content in the field of digital manufacturing,which provides more and more powerful support for the design and manufacturing of free-form surface parts.In this paper,we aim to improve the inspection accuracy and efficiency of free-form surface parts,the model simplification,point cloud description,coarse registration and fine registration problems are analyzed in detail.The mainly work of this paper are as follows.Firstly,because there are a large number of redundant points in the original scanning model,the directed Hausdorff distance method is introduced to solve the simplification problem.At the beginning,to define the feature points,the points were classified based on the curvature value.Then,the clustering method was employed for the subdivision of feature points.In addition,according to the region growing method,the directed Hausdorff distance-based technique was used to preserve the boundary point of the pint cloud effectively.Combined with the above two strategies,the proposed method can effectively preserve the feature and boundary points of the point cloud.Therefore,the simplified model can indicate the particulars of the real measurement model as far as possible,and it can meet the accuracy requirements while ensuring the model’s simplicity.Secondly,in general,the 3D coordinate information is not enough to describe each point effectively.In this paper,a novel high dimensional pint descriptor is proposed based on electric field properties.Considering that the 3D point cloud and electronic cloud have some similar properties,in this paper,the measurement point cloud is considered as an electronic cloud with electrostatic equilibrium state.The electric field force,electric field intensity,electric potential energy and electric charge density were introduced as the point feature descriptor with a high dimension.The designed descriptor can integrate the intrinsic properties of each point effectively,so as to provide a more accurate registration for matching operations.In addition,based on the setting of feature thresholds,we can select the specific attribute bits of the points and find the feature points of the model,so as to improve the precision of registration.Thirdly,aiming at the problem that the two point clouds with low coincidence degree are difficult to match,a new coarse registration method is proposed based on local coordinate system.In terms of the relation between the query point and its neighborhood,the local coordinate system is constructed by the principal component analysis(PCA)method,and the maximum density method is employed to compute the space transformation matrix.This method is simple and does not require repeated iterations.It can provide better initial solution for subsequent fine registration,and accelerates the whole matching process.Moreover,the method is insensitive to the initial position and noise point of the two cloud points,and can get a better matching result when the difference between the two models is large.In particular,the proposed method is suitable for multiple point clouds registration in 3D reconstruction.Fourthly,considering that the existing fine registration methods are sensitive to the initial position and outliers of the pint cloud,a new matching algorithm is proposed based on the enhanced fruit fly optimization algorithm.In terms of the essence of registration,the proposed method is analyzed deeply in the encoding,decoding,population initialization,objective function design,optimization strategy choosing,and individual ability optimization,etc.Therefore,the proposed algorithm has strong global search ability for the registration problem.This method can improve the running efficiency of the algorithm effectively while ensuring high precision.The standard data and noisy data were used for the simulation matching experiments with various complex positions,respectively.The results show that the designed method had high precision and robustness,and can overcome the shortcomings of the existing fine matching methods.Finally,combined with the proposed methods,the inspections of aeroengine blade and engine crankshaft are conducted as an example.Furthermore,the proposed methods are compared with other algorithms in precision,efficiency and versatility.The experimental results verify the effectiveness of the proposed method in machining applications,and it can effectively reduce the production cost and shorten the processing time.