Application of neural networks in feature recognition of mould reinforced plastic parts

Abstract

Feature recognition is an application dependant task, which has been mostly focused in production planning of machining process. It plays a fundamental role and usually is the first step in downstream activities concerning product development process such as design for manufacturing, design for assembly and process planning. This report presents a methodology to carry out recognition of design for manufacturing features of reinforced plastic components. A three-layer neural network system was created and trained using back-propagation-supervised learning to recognise nine of the most important design features related to this manufacturing process. Also, a methodology for pre-processing 3-D solid models such that geometrical and topological information of the part could be suitable as network input is presented. High performance of the net system was achieved on the recognition of the trained features as it was observed in several test parts.