Predicting compositions and properties of aluminum die casting alloys using artificial neural network

Abstract

Despite the large number of existing alloys and alloy databases, identifying proper alloys for specific applications still remains a challenge. In order to facilitate the selection and prediction of aluminum die casting alloys and their properties, an electronic database - ?i-Select-Al? - has been developed by the Advanced Casting Research Center (ACRC) and the North American Die Casting Association (NADCA). The key to the predictions is the determination of a relationship between alloy properties, chemical composition, and processing variables. Theoretically, these relationships can be ?accurately? determined using fundamental physical principles. However, in practice, the underlying mechanisms are not fully understood and difficult to be utilized. In this case, approximate empirical models are considered. In version 1.0 of the software trend equations have been generated. The nature of these trend equations limits the applicability and prediction ability of the software. In order to improve the prediction power; relationships based on an artificial neural network (ANN) were exploited in version 2.0. ANN has proven to be a highly flexible tool, suitable to treat multiple-input conditions and nonlinear phenomena with complex relationships between input and output variables. This article presents the working mechanisms, the programming, and the application of ANN in this project. The results show that ANN is a valuable modelling tool for predicting properties- fromcomposition and composition- from-properties for aluminum die casting alloys.