Research On Key Technologies Of Digital Joining Large-scale Mechanical Components Based On Visual Positioning And Tracking

Recently, digital assembly and joining has already been one of indispensable parts for manufacturing large scale mechanical components thanks to the increasing level of automatic design and manufacture for mechanical components within several engineering fields, such as aerospace, automobile, and ship, etc. In practice, several optical measuring equipments and devices can be used to assist large scale mechanical components to assembly and join automatically, fast and accurately. However, domestic research on digital assembly and joining is still relatively weak and far away from the advanced technology of developed countries.To satisfy the requirement of digital joining for large scale mechanical components in the manufacturing industry, this dissertation has been supported by both National Natural Science Foundation and Beijing Natural Science Foundation, which focuses on few key technologies and issues of digital joining for mechanical components based on visual positioning and tracking. There are5achievements as following:(1) A visual positioning and tracking system is built. In the process of digital joining mechanical components, the visual positioning and tracking system can be used to acquire entirely3D profile of every component, which can be divided into a front measuring subsystem and a rear tracking subsystem. The former is applied to obtain local intensive point cloud by color-encoded structured light; according to the principle of stereo vision, the latter needs to complete automatic and accurate registration among different groups of point cloud by tracking positions and orientations of front cameras.(2) Based on De Bruijn sequence, color-encoded structured light is further studied. In accordance with optical plane-line triangulation, a color horizontal stripes pattern has been encoded by a De Bruijn sequence of order3over5color symbols. To avoid the drawback of traditional RGB algorithm, a HSI decoding algorithm has been developed to separate color values from intensity values effectively for locating edges of stripes with sub-pixel accuracy. Experimental results have shown that:a) Results from HSI decoding algorithm are better than that of RGB algorithm by isolating color values from intensity values, which ensures that the accuracy will not be affected by both mutual disturbance; b) Compared with the color horizontal slits pattern, the color horizontal stripes pattern can acquire higher resolution with similar accuracy.(3) An error evaluating system for coordinate transformation is established. Aiming to coordinate transformation involved in digital joining for mechanical components, both calculating sets-testing sets error evaluating model and calculating sets-calculating sets error evaluating model are defined; and3evaluating approaches e.g. the coordinate error method, RMS error method and the error method of relative Euclidean distance are analyzed in details; besides, an invalid case of the third evaluating method is proved that once testing set has been determined, error parameters is always a constant no matter what variation calculating set does.(4) An efficient approach for planning geometrical distribution of markers is proposed to ensure that planning a reasonable distribution of markers as few as possible within a known3D space will satisfy a given accuracy requirement. The approach mainly contains the following aspects:according to the classification of markers, distribution parameters is described by internal parameter and external parameter. The former is composed of point number, the coordinates of reference point and the coordinate difference, and the latter is defined as the overlapping degree between calculating set and testing set; influence of geometrical distribution of markers to transformation accuracy is formulated; subsequently, a guideline for planning geometrical distribution of markers is generalized and summarized; consequently, an adaptive algorithm for planning geometrical distribution of markers is presented based on a constraint of centroid with curvature weighted coefficients. Both computer simulating experiment and practical experiment have been used to prove influence of distribution parameters of markers to transformation accuracy and invalidation of the error method of relative Euclidean distance. Moreover, the adaptive algorithm has been verified to be correct and valid by measuring controlling field experiment and3D point cloud simplification experiment. With significance and application value, the proposed approach is suitable to solve coordinate transformation involved in several practical engineering fields, such as geological survey, vision measurement, biomechanics, robotic navigation, and multi-sensor fusion, etc.(5) Registration between measuring point cloud and CAD model is researched. Similar to point number of measuring point cloud, a discrete model point cloud can be generated through uniformly segmenting CAD model represented by NURBS surface. Registration between both groups of point cloud should be divided into coarse registration and fine registration. In the first stage, a coarse registration algorithm is presented based on a consistent distance constraint and a curvature invariant characteristic of rigid transformation, where a curvature point set is chosen to be invariant to calculate a rotation matrix and a translation vector. Subsequently, a proved ICP fine registration algorithm based on grid quartering searching method is designed by determining a least distance between a point and its corresponding tangent plane. Both simulating experiment and practical experiment has illustrated that the coarse registration algorithm can achieve higher accuracy and provide reliable initial values for the fine registration.