Mechanism Retrieval in Conceptual Design Using ART1 Neural Network

Selecting appropriate mechanism types that meet design requirements is a critical problem often encountered in conceptual design of mechanical system. A novel approach to mechanism coding is presented at first, in which the features of motion function and function quality for mechanism can be expressed respectively with a list of binary vectors. A retrieval approach to mechanism type selection is then proposed using adaptive resonance theory (ART) neural network. Under this approach, sets of binary vectors representing all mechanisms are fed into an ART1 network to structure mechanism clusters and a proper number of reference mechanisms can be received after a binary vector representing design requirements is fed into the pre-grouped network. Compared with other retrieval system, by adjusting the value of vigilance parameter, the designer can obtain an optimal mechanism or several satisfactory mechanisms more easily in terms of his design intent using this approach.