Research On Robust Registration Methods In Quality Inspection Of Complex Free-Form Surface Parts

In order to meet the increasing and complex performance requirements of the equipments in the fields of aviation,navigation and automobile,the shape of manufactured parts becomes more and more complex.The advanced processing method and accurate quality inspection are essential to guarantee the accuracy of the complex free-form surface parts.3D point cloud registration is an important part in the fields of digital manufacturing and quality inspection,which provides technical support for the design and manufacturing of complex free-form surface parts.Therefore,to improve the inspection accuracy and robustness of complex free-form surface parts,the 3D reconstruction method of complex parts,the coarse registration and fine registration method between sampling points and design model are analyzed in detail.The mainly work of this paper are as follows:In 3D reconstruction methods of complex parts,the rigid registration problem is a6-DOF optimization problem.The heuristic algorithm must find optimal rotation and translation parameters to align adjacent measurement data.However the dimension of the feasible solution space is high.The setting of optimization space is very important for finding globally optimal solution,and manual intervention is required.Therefore,a heuristic registration method based on 3D rotation space optimization is proposed in this paper.Because the normal vector is invariant to translational motion,the closeness between point correspondences is determined by their normal vectors.Considering the correspondences degeneration cased by noise and occlusion,the Pauta criterion is introduced to eliminate abnormal matches.Therefore the flower pollination algorithm only needs to find optimal rotation parameters in searching space.The experimental results demonstrate that the proposed method has better registration accuracy than other local optimization and heuristic methods.Considering that the initial position between the sampling data and design model is complex and unknown,a coarse registration method,which is based on clustered medial-surface points,is developed.Firstly,the inner medial-surface points of objects are obtained based on the location of the points and their normal vectors.Due to the non-uniformity of medial points,the mean-shift algorithm is adopted to extract key-points which have higher point density than their neighborhood points.Above the initial matching based on the radius of inner circle,to reduce influence from symmetrical morphology,the feature correspondences are obtained based on the distance between feature points and their sequences on tetrahedrons.The experimental results demonstrate that,compared with other methods,the method proposed in this paper is more robust to different disturbances such as noise,density variation and overlapping ratio in most cases.In fine registration of complex parts,the traditional probability distribution method is still inefficient and sensitive to initial position.An alignment method based on Gaussian mixture model via bi-directional distance is proposed.In order to ensure both efficiency and registration accuracy,an improved fast Gaussian transformation is introduced,and the expansion center is modified based on segmentation.The partial differential relationship between the objective function and the rotation angle is considered to ensure the robustness to the initial position.To verify the robustness of the proposed method,the noise,outliers and initial position are tested.The experimental results demonstrate that the method is robust and efficient.For the registration method based on probability distribution,the optimal solution is obtained through the local optimization method.Thus the initial solution is still necessary.To deal with this problem,the heuristic method is introduced into solve different objective function based on different probability distribution.And their performances with regard to noise,overlapping ratio and density variation are discussed.Finally,the inspection of free-form surface and bearing support is taken as an example.In addition,the proposed method is compared with the traditional algorithm in terms of accuracy and versatility.The proposed method achieves good performances.Moreover,the effectiveness of the proposed method in the quality inspection of complex free-form surface parts in verified by the experimental results.