Machine learning based a priori prediction on powder samples of sintering-driven abnormal grain growth

Abstract

Reducing porosity in sintered products allows enhancement of several of its properties such as conductivity (electrical and thermal), strength, and translucency. One of the key challenges to porosity reduction and property improvement during sintering is abnormal grain growth (AGG). Abnormal grain growth occurs when a certain energetically favored particle becomes significantly larger than other particles in the matrix. This leads to an increase in porosity due to uneven particle size. There is no simple a priori test to determine whether a given powder sample will exhibit abnormal grain growth, when subjected to sintering. Here we show that a machine learning based approach predicts abnormal grain growth in powdered samples prior to actual sintering. This approach has the potential to allow for pre-selection of appropriate powder samples with an accuracy of 82%, to minimize the risk of abnormal grain growth in practical sintering processes.