Aluminum alloys based on alternative alloying systems, particularly Al–Ca eutectic alloys, offer a favorable combination of castability, mechanical performance, and corrosion resistance. A key step in developing such alloys is reducing the time and labor inputs required to obtain a promising composition with the desired set of properties, a task partially addressed through thermodynamic modeling software. In this article, various machine learning methods, including linear regression, decision trees, and two types of gradient boosting, are applied to optimize alloys within a composition–property relationship framework. The database comprised 250 cast aluminum alloys based on Al–(Si, Ca, Ni, REM, Mg, Cu) systems described in scientific publications, with data on chemical composition, hot tearing susceptibility, hardness, and strength. The most effective method for predicting the mechanical and casting properties of as-cast alloys was identified by comparing predicted and experimental values on a hold-out dataset. The best results were achieved with Yandex’s CatBoost gradient boosting model, built on open-source code and modified for the purposes of this study. Differences between predicted and measured values are illustrated using alumocalcium alloys as an example.
扫码关注我们
求助内容:
应助结果提醒方式:
