Model preparation begins with validation of the input model to ensure it is a solid; if it is not, it must be repaired. (Models are corrupted either by designer misunderstandings or inadequate CAD post-processing, as described below.) The valid model is then scaled and oriented with respect to the build chamber, taking into account build direction, build time, surface quality, and potential distortion. Many models may either be merged into a one-build assembly or nested for efficient utilization of the machine and material. The models may need to be compensated to account for downstream shrinkage or deformation. Finally, the supporting structures for overhanging part geometry as well as internal supports are added, if the process requires them. Such structures are often generated automatically in a separate file that is merged with the model file prior to slicing. The generation of control signals starts with slicing the model(s) and then scanning each slice into lines to determine the peripheral contour boundary needed to control the solidification process.
Model preparation consists of three steps: (1) validation and repair; (2) compensation; and (3) support structure generation, as illustrated in Fig. 8.5.
Any sampling process can introduce geometrical anomalies. For example, if spatial sampling is done too grossly, then small local features and local surface curvatures are lost irrevocably, or if two intersecting surfaces have significant differences in curvature, then problems can occur along their intersection curve. Difficulties in tessellation lie in lack of attention to detail. Many special cases occur when dealing with very complex surface shapes and their intersection and trim curves. Many pathological numerical idiosyncrasies also occur when dealing with tens or hundreds of thousands of very small and possibly ill-conditioned triangles. Some of these conditions are outlined in Fadel and Kirschman (1995), Mandorli and Cugini (1995), and Otto et al. (1995a).
Fig. 8.6 illustrates how triangular facets are sometimes omitted by tessellating routines, creating gaps or holes in the resulting model. If Surface 1 is relatively flat, then it can be approximated fairly accurately with large triangles, defining a series of widely spaced points along the curve boundary with Surface 2. If, on the other hand, Surface 2 has high curvature, then it must be approximated by smaller triangles, thus defining a series of more closely spaced points along the common curve boundary with Surface 1. When each surface's set of distinct points along the common boundary are connected by straight lines to create triangles, gaps or missing facets will appear (shaded). This situation can be easily corrected, but it is typical of a whole multitude of anomalies that require detailed attention.
A hole in the boundary surface of a 3D model will cause the specific slice that intersects the hole to have an incomplete -- not closed -- contour boundary. This condition, in turn, generates an erroneous control signal to a laser beam or other solidifying agent, causing the agent to continue solidifying material until it reaches the wall of the material container, thus fabricating a thin ray of material emanating from the part and producing waste.
Anomalies sometimes occur because of the shortcomings of the STL data format. Although the format is simple, there is much redundancy in the data. The STL format defines each triangle independently, by its three vertices and its outward normal vector. But in any tessellation, every edge and vertex of adjacent triangles are coincident, as Fig. 8.6 shows. Since the file format does not account for this fact, numerical error can cause edges and vertices that are supposed to be coincident to have different values. Generally, one can preprocess an STL file and add topology adjacency information, a posteriori, by testing edges and vertices for sameness and adding pointers in the file. Again, numerical errors can creep in. Topology information is most reliably obtained during tessellation. Introducing topology information simplifies the slicing routines and speeds up execution (Rock 1991).
Finally, anomalies occur because designers do not always understand the physical processing implications of their choice of geometric elements, or their placement or manipulation. Solid models have sometimes been constructed by abutting two solids together, introducing coincident surfaces in the model, as shown in Fig. 8.6. In the CAD world, this is not a problem because the coincident surfaces have zero thickness. However, their presence changes the way the material is solidified, leading in some cases to internal stress buildup and ultimate distortion of the physical part.
Various vendors have developed evaluation and repair software that determines if any triangles are missing in the tessellated model and fills the gaps with new triangles. Gaps consisting of a single missing triangle are easy to repair. The problem becomes more difficult when the gap consists of, say, 40 vertices. In this case, there are many different sets of triangles that could fill the hole. Unfortunately, the original sampled surface values needed to make that decision are not available. One option for creating the triangle-filling repair is to extrapolate the local surface curvature of known neighboring vertices. Ultimately, it is a matter of trade-offs. More complex solutions make sense only if high accuracy is achievable. It is much better to eliminate the problem up front. Various repair issues are discussed in Bohn (1993).
Table 8.4 lists the evaluation and repair capabilities of vendors visited by the JTEC/WTEC panel. D-MEC's system automatically repairs small cracks but requires manual repair of large cracks. CMET's STL file repair is topology-based. Presumably, CMET is checking for common vertices and edges, as discussed previously. The Denken/Autostrade Solid Laser Plotter software displays damaged surfaces and has the capability for extensive slice manipulation.
In Europe, Cubital representatives maintain that their Solider Data Front End (DFE) system has extensive model manipulation capability and can repair everything: it can cut, patch, or trim facets. Daimler Benz attacked the problem up front, which is possible for an internal system, by creating its own CAD support tools to generate error-free STL models.
Since physical parts shrink and deform under processing, models are needed to anticipate and compensate for these shape changes. Most users today follow a trial-and-error procedure, iterating through several trial runs before achieving the desired part. The JTEC/WTEC panel saw a number of studies underway, summarized in Table 8.5, that are aimed at replacing this intuitive process with one founded on formal analytical tools. The underlying difficulty in developing analytical tools is lack of detailed process understanding.
Research is underway at the University of Tokyo to simulate stress buildup and the resulting deformation. Denken Engineering/Autostrade is considering using finite element analysis (FEA) to create optimal scanning strategies that minimize distortion for specific geometries. This work requires the modeling of laser-polymer interactions to determine the resultant stress formation. Kira simply compensates 1.5-2% in the z direction to account for the swelling of its paper due to humidity. Nakamura Pattern Making Company considers high quality and precision to be the most important requirements, even more important than price. In fact, Nakamura bought a more expensive SFF machine rather than risk compromising these requirements with a less expensive machine.
Projects dealing with part distortion are underway in Europe as well. At the Fraunhofer Institute for Applied Materials Research (IFAM), designs are being optimized for stress using genetic algorithms, and the Institute for Polymer Testing and Polymer Science (IKP) at the University of Stuttgart is using finite element techniques to predict forces in multiple layers to better understand the resulting warp and curl.
Table 8.6 summarizes the current work that the JTEC/WTEC panel observed on development of support structures. The panel found that in Japan much of the software activity for generating support structures for RP parts is internal and proprietary. For example, CMET's customers design supports manually. While engineers at CMET are aware of existing automatic support-generator software that may suffice (see next section), they have decided to develop their own.
Concerning European activity for support structures, panelists saw very efficient capabilities under development. The scientists at EOS are working on optimizing the amount of material used (mesh style supports) in their Skin and Core Support software. MAGICS software from Materialise allows the building of lattice-like support structures that constitute only 10% of the part's resin volume and build time. Panel members understand that Laser 3D has some capability in designing support structures, but have no details. IKP is developing new hatching strategies for both sintering and stereolithography processes.
Several independent software houses have emerged in Europe over the past few years to support the growing RP market. The software activity in Japan seems to be in the service bureaus, like the medical software being developed by INCS, although the Ricoh SolidDesign solid modeler is an excellent starting point for developing an RP capability. The United States has many small RP software houses that cover all aspects of the market. Some software companies have extended their CAD and model creation capabilities, like Imageware in the United States. Others provide software for RP-specific operations like STL data viewing, data transfer, data models, or support generation, with the goal of independence from vendors. Materialise in Europe and Brock Rooney and Associates (1992) in the United States are typical of this latter category. The establishment of newer RP software companies, such as DeskArtes in Finland (http://www.deskartes.fi/rt.html), indicates that the market in Europe will continue its rapid growth. See Wohlers at http://lamar.colostate.edu/~wohlers/ for market data.
Research institutes such as BIBA (http://www.biba.uni-bremen.de/groups/rp/rp_page.html) have also developed a variety of interfaces, from a VDA-to-STL translator to new software that combines the point clouds from a multiviewed scanned object into a contoured model in the CLI layer format.
Software offerings of Materialise are representative of the excellent functionality available in Europe (http://www.materialise.com/). Materialise offers three major RP packages, whose specific capabilities are shown in Tables 8.7 and 8.8 and Fig. 8.7:
The CT-Modeler package provides a complete interface from a CT medical scanner to CAD systems or RP machines, as shown in Fig. 8.7. MIMICS is the medical front end of the package, in which segmentation of structures is done using 3D selection and editing tools. A 3D color image is generated from slice data. C-SUP generates support structures. Perforated support structures build four times faster, consume much less material, and are easier to clean than conventional supports. CTM interpolates slice data across layers and interfaces directly to RP machines. It can handle high-order interpolation algorithms. MedCAD takes the medical scan data and creates surface files that are directly usable for the design of custom prostheses in CAD systems. It supports an IGES interface.
Fig. 8.7. CT-Modeler functions.