Recognising 2.5D manufacturing feature using neural network

This study develops a neural network based methodology for recognising Manufacturing Feature (MF) using the Boundary representation (B-rep) information. The methodology is capable of recognising both basic MF (including standard MF and non-standard MF) and interacting MF. Firstly, both the edges` convex concave attribute and the edges` position (whether an edge belongs to the inner edge loop or outer edge loop), which reflects the edges` characteristics and the relationship between the bottom profile and its adjacent faces, were analysed to define the input vector for the neural network. Based on this, a BP neural network with a single hidden layer which contains ten neurons was obtained. The basic MF is divided into four groups, and can be recognised easily using the neural network. For the interacting MFs, a basic MF is recognised firstly, the bottom profile for the interactive MF is updated, and the interactive MF can be recognised.