Automating tolerance charting using graph theory

This paper outlines an applied research programme to automate tolerance charting using graph theory. The programme consists of two main phases: 1) developing a rooted tree technique for tolerance charting; and 2) developing parametrized tolerance charts. The first phase, which has been completed, involves the development of a rooted tree technique for producing tolerance charts and its implementation in an industrial software package called CATCH. The rooted tree technique is explained and illustrated with a simple example. The second phase, which has recently commenced, aims at trying to parametrize tolerance charts. The output of the second phase is intended to be a neural network that can accept the part geometry as input and output a "skeleton rooted tree". A preliminary project has been carried out to determine the feasibility of classifying trees with a neural network according to industrial practice in one manufacturing company. The results of the feasibility study are presented.