Grouping parts with a neural network

Recognition of objects is used for identification, classification, verification, and inspection tasks in manufacturing. Neural networks are well suited for this application. In this paper, an application of a back-propagation neural network for the grouping of parts is presented. The back-propagation neural network is provided with binary images describing geometric part shapes, and it generates part families. To decrease the chance of reaching a local optimum and to speed up the computation process, three parameters-bias, momentum, and learning rate-are taken into consideration. The contribution of this paper is in design of a neuro-based system to group parts. The network groups all the training and testing parts into part families with perfect accuracy. Performance of the system has been tested on a benchmark example and then by experimenting with 60 parts.