|Budget Amount *help
¥2,200,000 (Direct Cost : ¥2,200,000)
Fiscal Year 1992 : ¥700,000 (Direct Cost : ¥700,000)
Fiscal Year 1991 : ¥700,000 (Direct Cost : ¥700,000)
Fiscal Year 1990 : ¥800,000 (Direct Cost : ¥800,000)
The purpose of this research is to find a way in which we have a machine learn to assemble or design mechanical parts. In this research, acquiring knowledge from a mechanical assembly text has been experimented. In every text, assembly methods are shown using both explanation and illustration. Some assembly is too simple to require no explanation, but they are necessary for understanding both features of mechanical parts and the order of an assembly. For this reason, 4 cases have been experimented, which are (1) interpreting explanation under the supposition that models of parts are given, (2) synthesizing models of parts from an illustration, (3) planning assembly procedure from an illustration with explanatory phrases, (4) understanding explanations and illustrations by referring to one another. With any one of these approaches, several incomplete plans of assembly have been acquired. From the castel, the system successfully acquire a word sense like "start", which is often used in a sentence, "Start a setscrew into the mounting holes". The system without this meaning would fasten the screw or could not infer a length of insertion. From the case2,3 dimensional of cylindrical parts are created, and attributive models of objects are driven. They contribute to identify the real object corresponding to one of the models obtained. From the case3,4, by finding correspondence between parts and figure elements, the unique order of assembly is acquired.