Rational design of chemically complex metallic glasses by hybrid modeling guided machine learning

The compositional design of metallic glasses (MGs) is a long-standing issue in materials science and engineering. However, traditional experimental approaches based on empirical rules are time consuming with a low efficiency. In this work, we successfully developed a hybrid machine learning (ML) model to address this fundamental issue based on a database containing ~5000 different compositions of metallic glasses (either bulk or ribbon) reported since 1960s. Unlike the prior works relying on empirical parameters for featurization of data, we designed modeling guided data descriptors in line with the recent theoretical models on amorphization in chemically complex alloys for the development of the hybrid classification-regression ML algorithms. Our hybrid ML modeling was validated both numerically and experimentally. Most importantly, it enabled the discovery of MGs (either bulk or ribbon) through the ML-aided deep search of a multitude of quaternary to scenery alloy compositions. The computational framework herein established is expected to accelerate the design of MG compositions and expand their applications by probing the complex and multi-dimensional compositional space that has never been explored before.