Research On Point Set Matching For Ship Block Manufacturing

Application of digital technology has become a hot research point in ship industry of modern shipbuilding mode,and it brings new development modes and technical supports to the ship design manufacture and automation.The implement of digital shipbuilding,which aims at reducing the construction cost,improving the efficiency and the quality of ship construction,has become an important approach for improving the core competitiveness of ship enterprises in the international shipping market.As one of the key technologies of shipbuilding,accuracy management and control of ship manufacture is a research on the construction precision,which runs throughout the whole life cycle of shipbuilding.Comprehensive reinforcement of the accuracy management and control of the whole process of ship manufacture is the key to ensure the structural quality of ship construction and the quick erection.Point matching technique has attracted special attention in the digital shipbuilding.It is one of the key techniques of accuracy management and control in shipbuilding,which is a fundamental and key technology in the field of computer vision and pattern recognition,including model reconstruction,image registration,error evaluation,workpiece localization,etc.The total station measuring system is used to construction precision analysis of hull block,there is no difference in all registration points for treating that makes the measurement data approach to the process model with the best square.This kind of the equilibrium error characteristic is suitable for the uniform distribution of geometric errors,and without considering the specific construction process requirements and the effect of the gross errors on the positioning accuracy,which may lead to the misjudgment of the detection accuracy to hull block.Therefore,the aforementioned problems need to be studied in order to obtain the best matching pose of hull block.The application of 3D scanning technology in shipbuilding is still in exploring stage,there are many mismatches between the pre-processing scan data and the corresponding process model.Conventional registration methods and data processing software matching would affect the positioning accuracy and cannot evaluation of construction precision due to the existence of gross errors and mismatching.To obtain the actual manufacturing errors of hull block and provide a reasonable evaluation parameter for construction precision,studies focus on issues of point set registration problem involved in the 3D scanning measurement system.The main research works are summarized as follows:1)When there was no reliable positioning datum or no datum in the process of hull block assembly,erection,in consideration of the assembly and welding sequence of feature section and the regional difference in precision,this thesis proposed a registering and positioning algorithm of multifeature considering the regional difference in precision.This algorithm was based on the theory of configuration space of multifeature,which utilized the sensitivity relationship between the registration error and the micro perturbation of the transformation parameters,and integrated the orthogonal basis projection of the registration transformation and specifics components zero setting method to treat the influence strength of the feature surface data with different precisions in regional precision differently.Firstly,the high precision regional data was registered,then the sensitivity of the parameters of this registration transformation was analyzed and registration constrains on some principal component directions in the transformation space determined.Finally,by means of the orthogonal basis projection of the transformation and specific components zero setting method,the secondary precision regional data could only transform in the unconstrained directions.The results showed that inducing the constraint of principal component direction into multifeature model registration can improve the positioning accuracy on secondary precision regional data on the premise of guaranteeing high positioning accuracy on high precision regions,which was suitable for actual requirement in engineering.2)To solve the problem that the measuring system of total station cannot effectively suppress the influence of gross errors on the positioning accuracy,a robust registration algorithm based on degradation was proposed to weaken the influence of gross errors on the registering pose by improving M-estimation function,and the convergence of the established robust model was analyzed.According to the M-estimation criteria,the algorithm obtained the registering weight coefficients of different precision data,and used a degradation iterative method to establish a robust estimation model to improve the reliability and convergence speed of the positioning solution.The effectiveness of the algorithm verified by both theoretical and hull block experimental results,which provided a quantitative data basis for the verification of models and the subsequent assembly.3)Pose registration between the 3D scanning data and the theoretical model is the key to the management and control of ship construction precision.It is difficult for conventional registration algorithms to obtain robust registration results,in order to realize fast and accurate registration,a robust point clouds registration algorithm based on maximum correntropy criterion was presented to improve the registration accuracy and actively promoted application of the 3D scanning technology in the field of shipbuilding.Core of the proposed algorithm was Gaussian kernel function and was introduced into registration learning in this study.A robust estimation function with Gaussian kernel function and a linear solution system based on hypothesis theory were built,and the objective was optimized by using an iterative strategy to test and adjust the incidence matrix.Theoretical and experimental results showed that the weight of gross errors can be reduced adaptively during the registration process,and the impact of gross errors on the position accuracy can be reduced as far as possible.4)Due to the complexity of 3D scanning data of hull block and the performance limitation of rigid registration algorithms,rigid transformation usually cannot handle effectively degradations,including noise,nonrigid deformation and outliers.In order to improve the registration and positioning accuracy of 3D scanning data for hull block,a coregularization robust nonrigid point set registration algorithm was proposed by using Gaussian mixture model(GMM)and regularization technique,which aims to estimate the potential correspondences and recover the underlying transformation between two given point sets in a unified GMM-based probabilistic framework.The registration problem was formulated as a mixture model estimation problem by employing two topologically complementary constraints in a GMM-based learning framework.The first is a Tikhonov-based regularization constraint that maintained global spatial motion consistency,and the second was manifold regularization constrain that preserved the intrinsic structure of a point set.Research on nonrigid point set registration algorithm can provide technical support for the application of 3D scanning technology in the field of shipbuilding.