1.0 INTRODUCTION
As part of a collaborative research effort, General Motors (GM) and the EDRC set a goal of producing CAD/CAM applications to reduce by nearly an order of magnitude the overall time from the design of a part to its manufacture, while additionally providing an object of better quality. To achieve this goal, research focussed on two areas: computer-aided design evaluation and rapid prototyping. GM's Inland Fisher Guide division worked closely with the EDRC to engineer and apply technology useful in their design and manufacture of plastic injection molded parts.
The EDRC developed the non-manifold geometric solid modeling system of Noodles to underlie and unify the CAD/CAM research. Noodles provides a single application environment for design, analysis, and reasoning about solid objects in one, two, and three dimensions. For design, it provides Constructive Solid Geometry operations for interactive model creation and a link to relational databases for non-geometric (attribute) information.
Research in design evaluation centered on providing critiques of a Noodles model in terms of potential manufacturing or assembly problems. Technologies developed using the Noodles system included shape abstraction and feature recognition. The parameters of recognized features were then used in a knowledge-based evaluation of manufacturability. The intent was for problems to be identified and designed out of a part well before its first physical prototype was made. Given that designs were already routinely rendered within CAD systems, such analysis was the next logical step in the use of computers in design.
Rapid prototyping made use of a commercial product, the Stereo Lithography Apparatus (SLA), which allowed creation of a physical, polymer-based prototype virtually over night. There was a necessity, however, of transforming the GM CAD/CAM models into the representation required by the SLA, which the Noodles system provides. Polygonization technology was invented by the EDRC to transform the surface-based representations provided by GM's system to a linearized, Noodles model. The SLA capabilities, as well as feature-evaluation technologies, were thereby linked to the GM system. GM has thus been able to decrease the time it takes to produce a first prototype, from an average of six weeks (with some prototypes having taken months) to a just a few days.
A knowledge-based approach to manufacturability analysis relies upon rules representing design standards and a proven base of human expertise. These rules generally work with abstractions known as features, which can be extracted from information available in conventional CAD systems. Two approaches to extracting features from a geometric model are presented. One approach uses a Differential Depth Perception Filter to select elements of interest. The other uses the results of three-dimensional thinning process known as the Skeletal Transformation to approximate a Medial Axis Transform. Both methods address issues in geometric and topological variation and seek to avoid the combinatorial explosion problem involved in an exhaustive search of a geometric model. However, these systems still do not eliminate the possibility of intractability with large scale models because some form of graph-matching is involved and thus incorporates issues of NP-completeness.
Technologies for feature extraction and knowledge-based analysis are developed in the broader context of a concurrent engineering environment; results of these software utilities should be readily available within this environment. Four common issues have arisen out of the development of these software utilities. These include 1) the necessity to reduce the problem space by filtering, transformation, or abstraction, 2) the discrimination and classification of features, 3) the mapping of features back to the original part, 4) and the evaluation of feature attributes.
Further research described in this thesis was based on a dismantling of the design environment to its basic components, and then a rebuilding of these components using the sign systems of semiotics as reference. Plastic injection molding design was an excellent domain for exploration of broader issues of design and modeling. Plastic injection molding uses the experience of molding experts to design molds while there are also well-developed numerical tools to analyze the flow and cooling of molten plastic within the mold. As such, both preliminary and formal design considerations are used.
Section 2.0 starts with an overview of plastic injection molding and includes a study on design changes that reflect the types of preventable design deficiencies typically found during manufacturing and assembly. The common ground between experience and theory is to be structured by the Analysis, Synthesis, and Evaluation (ASE) model of the design cycle. This section describes this model and its use as a tool for exploration of the elements of the design process. Object-oriented design, semiotics, and critical theory are related as considerations of models for computers.
Section 3.0 provides an overview of the GM Design For Molding System. Section 4.0 focuses on the Plastic Injection Molding Expert System (PIMES) Design Advisor. Here the extraction and use of features for heuristic analysis is discussed, as well as the particular implementation of this advising agent. The application of heuristics represents the experience of an expert in the molding field. This is an example of one stage of preliminary design, where the common ground between experiential and formal models is exploited. While there is a degree of automation in the characterization of basic features, there are some shortcomings. And while a design-with-features approach is clearly desirable, there are the unavoidable possibilities for the creation of secondary features and concerns as these features are added into the global context of the background.
Section 5.0 reviews the implementation of the Design Processor, an application that managed the files and running agents of the DFM System. Deconstruction of the information flow model to a more manageable artifact-centered model is illustrated. Details of the requirements for an engineering environment consisting of cooperative, interacting agents is also described. The visualization component of the DFM system is detailed in this section. The concepts of semiotic iconic-objects, particular in terms of the CLIP model, are illustrated in this section.
The construction of models, as illustrated by the definition of complex features, raises another set of issues that are to be explored in Section 6.0. Starting with the basic elements of nodes and links, objects and models are defined. Morphology, syntax, and semantics are discussed. Semantic difficulties play a large role in modeling difficulties. As an example of model building, a complex feature modeling language will be built, and issues surrounding the model and its modeling language will be explored. The similar nature of rules, tools, and types will be explored as residing in the dimension between experiential and rationalist views of the world. Conclusions and future work are discussed in Section 7.0. Of interest is the formulation of theory and design and the role of context in this formulation.