Multilayer artificial intelligence for thermal-conductivity prediction of silicon nitride ceramics from powder processing conditions and predicted densities

In this study, we first developed an artificial intelligence (AI) that estimates relative densities (RD) of silicon nitride ceramics from the process conditions. We then constructed a multi-layer AI that predicts the thermal conductivities (TC) from the above process conditions using an RD obtained by the developed AI. The RDpredictive AI utilized input data (explanatory variables) which represent the effects of main powders, sintering additives, organic sacrificial pore formers and heat treatments (nitriding and/or sintering), whereas the TCpredictive AI exploited the predicted RD as well as the aforementioned explanatory variables. Both the AIs successfully improved the prediction accuracy by incorporating the types and concentrations of sintering additives and conditions into the explanatory variables. These AIs exhibit potential to predict even other properties of silicon nitride ceramics prior to the real fabrication.