Configuration Design With Neural Networks

Abstract

Abstract This research examines a novel application of artificial neural networks (ANN) in engineering design. A NN-based learning technique is presented for generating configuration designs from functional requirements and basic domain concepts. With functional requirements as input, an elementary neural network can learn simple domain design rules. The output of the network is a set of complex design functions, later transformed into configurations. The application of design rules provides a strategy to enable an efficient search for acceptable design configurations. We validate the methodology in the domain of fluid power systems by a circuit simulation and analysis package Bathfp. This research is creating new insights into the practical realisation of an automatic configuration design process. It also sheds light on novel applications of neural networks.