EUROPEAN CONGRESS AND EXHIBITION ON ADVANCED MATERIALS AND PROCESSES - EUROMAT 2021

A traditional design alloy is mainly based on a ‘trial-and-error’ process that needs lots of time, effort, and cost. A more effective approach to design alloys using computational models has attracted much attention as computational technologies development. In this work, we applied computational approaches to the design of Ni-based amorphous alloys effectively in order to propose the novel alloy composition that has appropriate physical properties and better economy. Machine learning algorithms were used to design a new Ni-based amorphous alloy based on the analysis of literature data. The random forest regression (RFR) was chosen because of the best performance to predict thermal properties. At the same time, particle swarm optimization (PSO), one of the inverse design method, were considered to investigate an optimized alloy composition which shows target properties. The main thermal properties of Ni-based amorphous alloys such as glass transition temperature and crystallization temperature could be predicted by RFR and PSO. The influence of alloying elements on the thermal properties was quantitatively analyzed. The optimal alloy composition was achieved based on the predicted thermal properties and validated the performance with experimental results. We suggested an empirical equation to readily calculate the thermal properties without complicated machine learning modeling. We proposed a computational alloy design method for novel Ni-based amorphous alloys and anticipated that this approach could be adopted to predict other properties of Ni-based amorphous alloys and navigate new composition of various series of amorphous alloys.