Reverse engineering is a rapidly evolving discipline, which covers a multitude of activities. In this paper we will only be concerned with reverse engineering of shape, but a broader interpretation of the term to involve understanding of design intents and mechanisms is also possible. While conventional engineering transforms engineering concepts and models into real parts, in reverse engineering real parts are transformed into engineering models and concepts. The advantages of the extensive use of CAD/CAM systems need not be reiterated here. The existence of a computer model provides enormous gains in improving the quality and e ciency of design, manufacture and analysis. Reverse engineering typically starts with measuring an existing object so that a surface or solid model can be deduced in order to exploit the advantages of CAD/CAM technologies. There are several application areas of reverse engineering. It is often necessary to produce a copy of a part, when no original drawings or documentation are available. In other cases we may want to re-engineer an existing part, when analysis and modi cations are required to construct a new improved product. In areas where aesthetic design is particularly important such as in the automobile industry, real scale wood or clay models are needed because stylists often rely more on evaluating real 3D objects than on viewing projections of objects on high resolution 2D screens at reduced scale. Another important area of application is to generate custom ts to human surfaces, for mating parts such as helmets, space suits or prostheses. It seems important to clearly distinguish between the concepts of a 3D copier and a 3D scanner. A photocopier takes a piece of paper and produces another piece of paper just like the original. A 3D copier is a device which takes a solid object and makes another one of just the same shape (let us ignore material). In fact, copy machining has been a well established technology for a long time. A scanner however, in 2D, not only inputs a page of text into the computer, but can also recognize the characters and gures, thus providing a text le and graphical structures. Similarly, a 3D scanner will not only capture raw data from the object, but the data will be interpreted and some computer model will be created. Now, not only may a single copy be generated, but knowledge of the shape is obtained, and thus we can derive new shapes, make variations, analyze properties and determine characteristic quantities such as volume or surface area. The ultimate goal of reverse engineering systems is to realize an intelligent 3D scanner. However, there is a long way to go. Even capturing shape and translating it into a CAD model is a di cult and complex problem. In spite of several encouraging partial results in particular areas, a fully automatic solution to build a complete and consistent CAD model is still a goal. The purpose of this paper is to describe the most important elements of a reverse engineering system and to identify problems, which still require further research. At the same time, we attempt to summarize the basic achievements of current reverse engineering research, as well. The reverse engineering procedure can be characterized by the owchart in Figure 1:
Of course this sequence is fairly notional. In fact, these phases are often overlapping and instead of the sequential process shown, several iterations are required. Nevertheless, this outline may help the reader to understand the information flow and serves as a basis for organizing the content of our paper. A crucial part of reverse engineering is data acquisition. After reviewing the most important measuring techniques, the relative merits and di culties associated with these methods are discussed. Often, methods for reverse engineering are developedbased on simulated data acquisition only. Our experience is that a certain amount of reservation is needed in such cases, as actual physical measurements may display many problems and undesirable side e ects not present in arti cial data. As was indicated earlier, the main topic of this paper is the geometric part of reverse engineering. Data structures for representing shape can vary from point clouds to complete boundary representation models. We give later a hierarchy of shape models. This is particularly important since the representation chosen fundamentally determines the computational algorithms applied to the data sets. The most critical part of reverse engineering is segmentation and surface t- ting. By means of these processes, data points are grouped into sets to which an appropriate single surface can be tted. We believe that segmentation and surface tting methods must be carefully matched to each other. A range of techniques and problems will be described in the following sections, including methods for various surface representations used in CAD ranging from planes and quadrics to composite free-form surfaces. Data acquisition systems are constrained by physical considerations to acquire data from a limited region of an object’s surface. Hence, multiple scans must be taken to completely measure a part. See the section on combining multiple views. The problems of creating geometric models will be discussed in the last section. There are various representations providing approximate or incomplete models which may be su cient for certain applications, such as computer vision, animation, collision checking etc. For CAD purposes this is not so. Here we restrict our scope of interest and concentrate on accurate and consistent boundary representation models, using standard surfaces acceptable by commercial CAD/CAM systems. Identifying sharp edges, adding blends, providing proper continuity where smooth connections are needed, tidying up the model and enforcing constraints are all part of the problem. Finally, in the conclusion we present our view of the current status of this technology and what are the most important research issues. This paper is an overview, so we do not attempt to describe individual topics in detail. We try to concentrate on important conceptual issues, while hopefully the reference list at the end will help readers to nd the most relevant research contributions.
Courtesy Tam as V arady, Ralph R. Martin, Jordan Cox z