Thermal conductivity prediction of sintered reaction bonded silicon nitride ceramics using a machine learning approach based on process conditions

Abstract

Thermal conductivity (TC) of sintered reaction bonded silicon nitride (SRBSN) ceramics was predicted from process conditions using support vector regression (SVR) as a machine learning algorithm. A total of ten explanatory variables were selected from the process conditions for building a SVR model. The determination coefficient (R2) in the testing results of the trained SVR model reached a satisfactory level (R2 = 0.8 on average) when more than 100 data points were used as the training data. The relative importance scores of the explanatory variables revealed that sintering conditions (temperature and time) had the largest contribution to the TC prediction while those of oxygen impurity content in the main starting powder, types and concentrations of sintering additives and nitriding conditions were also influential to a certain extent.