CAD is a great acronym that could mean many things. My personal favorite is computteraided disaster.1 This look into the future uses CAD in the sense of computer-aided design of 3D geometry data. I further restrict the scope to applications that produuc geometry defining the shape of physical products. I consider three different aspects of CAD’s future: n geometric modeling and visualization, n making geometry available to other applications, and n keeping the data under configuration control. Brief history CAD was one of the first computer graphics applicatioons Ivan Sutherland’s Sketchpad at MIT and the DAC-1 project at General Motors began in the early 1960s. Industry developed its own CAD applications, delivered on multiuser mainframes, in the 1960s and 1970s. In the 1980s turnkey systems bundled hardware and softwaare Current implementations separate hardware and software components. As a result, CAD software most often executes locally on powerful Unix or Wintel workstattion in a distributed environment. Even though engineering drawings were the dominant output through the 1980s, early design stages relied on a variety of unique curve and surface forms. Customized systems were used in aerospace for surface lofting (TX-95 at Boeing, CADD at McDonnell-Douglas, CADAM at Lockheed) and in automotive for surface fitting (Gordon surfaces at General Motors, Overhauser surfaces and Coons patches at Ford, Bezier surfaces at Renault). Curreen CAD applications rely on general geometric forms like nonuniform rational b-splines (NURBS). Solid modeling also started in the late 1960s and early 1970s, but from different roots. Larry Roberts worked at MIT to automatically identify solids from photogrraphs The Mathematics Application Group used combinatorial solid geometry to define targets for nuclear incident analysis and subsequently developed ray-traced rendering and solid modeling in Synthavisiion Other efforts (such as TIPS from Hokkaido Universsity Build-2 from Cambridge, PADL from University of Rochester) had limited industrial impact. Solid modeling is the preferred technique in the 1990s for defining 3D geometry. While modeling issues persiist geometry modification, integration, and data manageemen are now the key issues. Geometric modeling Geometry data represents points, curves, surfaces, and solids. The many descriptive forms used range from precise canonical definitions (circle, sphere, cone) to general parametric forms (Bezier, NURBS, multiresolutiion) Solids add a level of complexity when bounded with general surfaces because topology must be defined. A digital computer’s simplicity causes problems that will continue as long as our current silicon-based technollog dominates.2 The finite precision of floating-point numbers forces geometry algorithms, which rely on them, to converge to some epsilon rather than the absolute zero that math theory requires. As a result, differren algorithms can yield different results for geomettri operations (intersection, union, fitting) and geometric analyses (curve length, surface area, solid volume). New computing techniques (neural, quantum, optical) may support more reliable algorithms. CAD users often stress the outer limits of compute power, memory, and storage. As Moore’s law continues to yield faster and cheaper computers, users generate greater numbers of more accurate models to digitally assemble, view, and analyze. We still have a long way to go to compose an entire commercial airplane with curreen technology. The Boeing 777, the first commercial airplane designed with solids, requires more than a milliio parts consuming over 5 Gbytes of memory for displla only. Figure 1 shows how detailed current solid models are. Contrast the digital model with the photogrrap of a physical flight deck to understand the gap. Numerous projects have investigated graphical techniqque (sculpting, drag-and-drop, snap dragging, sketching) for construction and modification. Success has been limited because users still see a 2D screen and use input devices with poor resolution. Therefore, they often use geometric operations to maintain acceptable accuracy. Pseudo-3D techniques like virtual reality and holography will be applied in later design stages (for example, to simulate maintenance tasks) until accuracc and performance are improved. Once created, geometry data must be analyzed to ensure that shapes fit properly. As the number of parts grows, they can be inspected automatically or visually for gaps and overlaps. As products become large, computtin and rendering resource demands increase. 0272-1716/00/$10.00 © 2000 IEEE Vision 2000 34 January/February 2000 David J. Kasik Boeing Viewing the Future of CADModifying geometry for derivative products is currenntl arduous because thousands of entities may be affected by one change. Future products based on a more abstract paradigm (scripting languages, geometry featurres will improve both generation and modification because higher abstraction lets the user capture complex relationships as early as conceptual design. Knowledge capture and management efforts will continue gaining momentum in both academia and industry. Existing physical parts can be reverse engineered into accurate geometry models. Input data can be acquired via direct digitization, video, or scanning. Adequate compuut power and memory are becoming available to allow fitting algorithms to work with acceptable precision. Integration with the rest of the world Computer-aided engineering (CAE) generally relies on discretizing geometry; describing other part attributte and physical conditions; and then burning signifi-cant amounts of computation with finite-element, finite-difference, or other algorithms. CAE applications digitally test key static and dynamic characteristics of assemblies or individual product shapes. Determining the best way to manufacture a physical product from geometry data has traditionally relied on defining key manufacturing characteristics like fillets, chamfers, pockets, and blends. Development of direct and generative techniques from geometry data will furthhe improve computer-aided manufacturing (CAM). Future CAE and CAM will provide tighter feedback early in the design-build process to improve product shape more quickly. Knowledge about the preferred way to improve geometry from CAE and CAM results and collaborative work will shorten product design cycles. Because of differences in user preferences, fundamennta capabilities, and salesman moxie, multiple incompatible CAD systems will continue to exist. PDES/STEP has improved the robustness of geometry descriptions and translation accuracy over standards like IGES. However, moving data from one CAD application to another will continue to be a problem as long as there are different geometry algorithm implementations. Product data management The CAD world is dominated by extremely large amounts of data produced both during geometry modellin (for example, Boeing dedicates about 14 terabytes to commercial airplane geometry data storage) and by CAE and CAM applications. Data volume increases substanttiall for new versions because one change in geomettr can cause significant redesign and re-analysis. Technology like data warehouses and product data managers treat geometry as blobs. Finer granularity is needed to improve searching for shape characteristics. Scaleable solutions must handle the hundreds of terabyyte that store multiple versions of CAD, CAE, and CAM data. The longevity of geometry data is the most underapprecciate CAD problem. As digitally defined products age, the data that defines them must age gracefully. Imagine a product with a 30-year life expectancy. Computing technollog produces new software and hardware products yearly. Bringing forward all the data and revalidating it on a yearly basis is cost prohibitive. Keeping old systems around is also expensive and risky. This problem is gettiin bigger as more companies rely on geometry data. Long-term data retention is an area in which part reverse engineering might offer back door insurance to let companies reconstruct parts when data is unrecoveraable This, however, does not address product liabilitty which needs key design documentation. Summary The world of CAD for physical products will continuu to push the limits of available computing resources. Even with the fundamental limits of geometry algoritthms CAD will continue to increase in strategic importannce especially as digital designs age and are transformed into new generations of products. n References 1. P. Mellor, “CAD: Computer-Aided Disaster!” High Integritt Systems J., Vol. 1, No. 2, 1994. 2. C. Mead, “Life Without Bits,” in Talking Back to the Machine, Copernicus, New York, 1999. Contact David.j.kasik@boeing.com. IEEE Computer Graphics and Applications 35 1 Rendered geometry of a Boeing 777 flight deck (a) and photogrrap of a physical 777 flight deck (b). (a) (b)