Adaptive neuro-fuzzy inference system approach for tensile properties prediction of LPDC A357 aluminum alloy

Abstract

This study is based on the desire of aluminum casting foundries to understand the influence of minor changes, within the specification limits, in the alloy chemistry. In order to ensure the casting of A357 Al alloys within the framework of the casting standards and to minimize the quality problems that may arise during casting; the estimation of ultimate tensile strength (UTS), yield strength (YS) and elongation (ε) due to very small changes among the alloying elements, although they are in the standard range, by using machine learning method (ML), were studied. The dataset of chemical composition and tensile properties of Low-Pressure Die Cast (LPDC) A357 Al alloy were experimentally established. The relationship between five input variables in the A357 alloy, namely the main alloying elements Si and Mg together with the most common impurity contents Fe, Ti and Cu were selected and three outputs (i.e UTS, YS and ε) were linked by Adaptive Neuro Fuzzy Inference System (ANFIS). The ANFIS model predicted that the most detrimental element affecting tensile properties was Fe content. According to this model, the order of the relative importance on UTS, YS and ε revealed as Si, Mg and Ti content respectively after the Fe content of the alloy.