Feature Curve Reconstruction Of Large Forgings Based On3D Point Clouds Meshless Processing

Large forgings are the fundmental part of the major equipments. Precise extractionof the size of the large forgings plays important roles in improving the utilization of rawmaterials, improving product quality and increasing the rate of qualified products.Reconstruction the feature curve of the workpiece from measured3D point clouds isgradually becoming a hot and difficult point in the field of measurements of largeforgings size. Based on the strong noises, mass and localized characteristics of the3Dpoint clouds of large forgings size, utlizing3D point clouds meshless processing theory,the problem of accurate, rapid and reliable reconstruction methods of the feature curve oflarge forgings were researched indepth.Firstly, based on anisotropic denoising theory, a new3D point clouds anisotropicdenoising model was estiblashed for large forgings3D data, using the differentialgeometric information of the sampled points. The shape and orientation of the denoisingkernel function were ajusted adaptivly. A new3D point clouds denoising alghrithm wasproposed for large forgings based on the new model. The strong noises are smoothed,meanwhile the feature feature of the forgings are preserved, laying the foundation for theaccurate reconstruction of the feature curve of large forgings.Secondly, measure theory and estimation theory of discrete differential geometrywere combined together. The information similarity model was established toquantitativly describe the differential geometrical similarity between the the sampledpoints. The contract pairs were constructed under the princple of maximum informationsimilarity. The information similarity weighted quadric error matric was construced foreach sampled point based on the information similarity between the sampled point and itsneighbors. The optimal position of the contracted pair was solved by minmizing theweighted error matric, the cost of the contraction was evaluated by the minmum value ofthe weighted error matric. A3D point clouds meshless simplification algorithm wasproposed for larging forgings based on the concept of iteratively contracting the least costpairs, providing guarantees for the accurate and rapid structural curve reconstruction of large forgings.Thirdly, a new3D point clouds registration measue was proposed based on kerneldensity estimation theory. The BFGS quasi-Newton optimization and simulated annealingalgorithm was combined together to search the maxmum point of the measure in varyingscaled manner. A robust variant scale3D point clouds registration algorithm was proposedbased on the principle, in order to solve the conflicts between accuracy and convergenceinterval of classical registration algorithms, laying the foundation for the integratedreconstruction of the feature curve of large forgings. Tensor decomposition theory andmulti-scale analysis theory were combined together. Multi-scaled tensor characteristicindex was constructed to describe the characteristics of the sampled points. Based on thecapability of the Multi-scaled tensor characteristic index in distinguishing the featurepoints, the minmum spanning forest algorithm was usd to connect the feature points andconstruct the featured polygonal lines. Furthermore, polygonal lines were smoothed bymoving least squares algorithm, and the fake feature points were projected to thesmoothed curves. In such way, the featured curves of large forgings were constructedreliably.Finaly,3D measurement of the simulated forgings was undergone by the laserscanning3D measure platform. The experimental results reveals that the accuracy,rapidity and reliablity of the proposed methods.