Parametric design by learning

Abstract

Parametric design is an effective and productive tool for the definition and modification of geometric models. This paper presents an intelligent method for the creation of a parametric model. In the proposed method, an adaptive neural network model is built to map the set of dimensional parameters to a set of coordinates. Defining a parametric model is equivalent to teaching the neural network. The user needs only specify a set of dimensional parameters that defines the parametric model and teach the neural network how to react to the changes of the dimensional parameters. Once the neural net work is taught, any dimensional changes will result in corresponding coordinate changes. This novel method eliminate the need of programming or graphic interaction that are normally required by contemporary parametric design systems.