Classification of the Mechanisms of Liquid Metal Embrittlement Via Machine Learning

Predicting liquid metal embrittlement (LME) between arbitrary liquid–solid metal pairs is particularly challenging; conflicting reports in the literature have prevented the creation of generalized predictive models. Recent advances have shifted the LME paradigm from binary—embrittlement either occurs or does not occur—to multiple, potentially cooperative mechanisms. In this work, a dataset comprising the vast majority of LME experiments reported in the literature was created for training machine learning (ML) models to predict LME under the new paradigm. Several ML classification techniques, including k -nearest neighbors (KNN) and decision tree classifiers, were trained to predict the occurrence and mechanism of LME for a given liquid–solid pair. Specifically, a KNN 1-vs-1 classifier was > 80% accurate in predicting LME, surpassing any prior models in the literature. Additionally, the decision tree models shed light on the features most important for predicting LME and identifying the mechanisms.