Geometric reasoning for intelligent manufacturing

anufacturing automation is at least as old as the industrial revolution, which brought about mass production and dedicated, fixed-task machinery. Flexible automation and the associated variable-task, programmable tools date from the 1950s, with the introduction of numerically controlled (NC) machining and computer-aided design (CAD). A modern, high-level view of the automation field can be summarized as follows: At the center are representations, or computer models, of products, such as an assembly of electromechanical components; processes, such as a sequence of machining operations; and resources, such as machine tools or robots. Around this center are the activities that create and use the representations. Creating product representations is usually called design, or CAD when emphasizing its computational aspects. In industrial practice, a design for a mechanical part or assembly is specified by its geometry plus a few nongeometric properties, such as material or hardness. Geometry is traditionally conveyed through engineering drawings. CAD helped replace such drawings, first with wireframe models, or unorganized sets of object edges, and later with solid models. Only solid models are unambiguous, completely defining the shape of objects; other models are ambiguous. Therefore, solid models are suitable sources of data for programs that automatically answer such queries as: “What is the volume of this object?” and “Do these two objects collide?” The evolution of solid modeling has been documented in the literature [7, 8, 9, 10]. However, the conversion to solid modeling is far from complete; drawings and wireframes are still widely used in industry. Today, human users define a product’s geometry directly. In the future, mechanical design is likely to follow in the steps of very large-scale integration (VLSI) design and specify products at a higher, functional level. Geometry will then be inferred through the analogs of silicon compilers. We are still far from this stage; a major barrier is the lack of a computational characterization of design intent in terms of mechanical functions, constraints, optimization criteria, and perhaps other important properties. Design is closely linked to analysis. Computational methods for analyzing designs have progressed nicely, and today there are tools for computing mass properties of objects, detecting interferences, simulating motion, and so on. Progress on synthesis of product and process designs has been much slower. High-level process representations are created through activities usually called process planning and scheduling. Planners reason about the geometry (and other properties) of objects, attempting to match them to the capabilities of physical processes, such as drilling or milling, and thereby generating a sequence of production operations. These operations involve fabrication, such as machining or casting; assembly; inspection; and so on. Schedulers reason about processes and resources, allocate operations to appropriate machinery, and specify the asso-