1998 年 13 巻 2 号 p. 263-273
The final purpose of the research is to build an AI system which automatically understands assembly procedures from an assembly manual and carries out the assembling of repairing tasks. An assembly manual consists of explanation and illustration. Part names and assembly sequences are obtainable from explanation, and configuration of a part in assembly is depicted in illustration. To combine the particular two parts specified in explanation, the system has to find the portion from illustration in which each part is located. Consequently, a part name is first extracted from illustration, then the figure is found from illustration which corresponds to the 3 dimensional model of the part with the extacted name. Part names are identified using a pattern matching method free from unknown words which allows a variable to be matched to any word sequence. The figure is examined to see if it can be matched to the model by referring to an auxiliary line which helps us understand assembly relation after inferring a line of sight used when the illustration was described. The illustration being drawn by hand, the figure may not be judged to be equal to the model of itself. By considering the fact, the posture of a model is discovered from illustration using a simple but robust matching method. As the result, except for the special cases not satisfying the assumption on parts used, assembly operations will be successfully simulated by moving each part along the auxiliary line using the new illustration in which all parts are replaced with their corresponding models.