Direct layered manufacturing of point sampled objects

Abstract

Though both 3D digitisation and Layered Manufacturing (LM) are well-developed technologies, still much research has to be done in direct integration of the two. The work described here is an approach based on a neural network algorithm for contour reconstruction from unorganised point cloud data, for direct LM of point-sampled geometry. A new Sliced Point Data Representation (SPDR) model has been proposed for this integration. This process constitutes the following steps: first, the cloud data is subdivided into thin slices of uniform thickness. The thickness chosen is the minimum thickness manufacturable by a specific user specified LM process. Then the individual sliced points are projected along the build direction i.e. Z to yield 2D point sets. These 2D point sets are approximated by closed polygon conforming to a given tolerance limit. For this step we use a Growing Self Organising Map (GSOM) neural network algorithm developed by us for curve reconstruction from unorganised thick point data. It is computationally effective and efficient in extracting principal curve from unorganised thick point data. The point data for slices that are within the tolerance of the previous contour can be merged and no additional cross section contours need to be generated. This results in an adaptive contour model and is stored as SPDR. A one-way translator for translating this SPDR model to a slice format .ssl file that can be imported by the LM system (Stratasys FDM) in our case has been developed. Case studies are then presented to illustrate the efficacy of this approach.