A new strategy for long-term complex oxidation of MAX phases: Database generation and oxidation kinetic model establishment with aid of machine learning

Abstract

Owing to competitive behavior between oxidation products, complex oxidation commonly exists for MAX phases applied at high temperatures. Two major challenges remain to explain the oxidation law, i.e., acquirement of comprehensive oxidation data and establishment of reliable kinetic model. In this work, the long short-term memory recurrent neural network (LSTM-RNN) model is adopted combining the thermogravimetric (TG) experiment to generate the comprehensive oxidation database of MAX phases. By exploring the working principles of machine learning (ML) algorithms, a novel approach of combining real physical picture (RPP) model and sure independence screening and sparsifying operator (SISSO) method is proposed. The obtained machine learning-based real physical picture (ML-RPP) model can accurately deal with the long-term complex oxidation of various MAX phases. This work will provide a useful guideline for the cognition of complex oxidation of other ceramics and alloys.