Conversion of Scanned Engineering Drawings to Product Data Descriptions

Despite the popularity of CAD/CAM systems today, over 90% of all drawings in the U.S. are not computer processable as they exist only in paper or microfilm. Without the development of technology to convert raster drawing to product data model, this situation is like to persist for some tine as over 50% of all new drawings are still being executed on paper. This vast backlog of manual drawings is a serious problem to logistic support as most engineering projects involve incremental changes to, or reuse of components in existing designs. The current practice of overlaying vector drawings on top of raster images necessitates investment in vast amounts of persistent data storage. Raster images cannot be used with modern CAD/CAM/CAE systems. This proposal describes a system that will input information from a raster to vector converter, and with human assistance, output produce-data models (PDES). The system uses an approach that takes advantage connectionist techniques (Neural nets and Fuzzy logic) with traditional knowledge-based techniques. A three-phase process is used: 1) recognition of abstract topological patterns, 2) hypothesizing product data prototypes for these abstract patterns, and 3) validation of prototype shapes within the context of the evolving product-data model using constraint propagation.